File size: 53,308 Bytes
28677d8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
[
  {
    "id": "docs-ollama-models",
    "certification": "Docs Study",
    "title": "Ollama ๋ชจ๋ธ ์‹คํ–‰ ํ๋ฆ„",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-1", "docs-lab-2"],
    "summary": "Ollama๋Š” ๋กœ์ปฌ์—์„œ LLM ๋ชจ๋ธ์„ ๋‚ด๋ ค๋ฐ›๊ณ  ์‹คํ–‰ํ•ด CLI ๋˜๋Š” REST API๋กœ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ด์ฃผ๋Š” ๊ฒฝ๋Ÿ‰ ๋ชจ๋ธ ์„œ๋ฒ„์ž…๋‹ˆ๋‹ค. `ollama run`์œผ๋กœ ๋Œ€ํ™”ํ•˜๊ณ , `ollama serve`๋กœ API ์„œ๋ฒ„๋ฅผ ์œ ์ง€ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "# ๋ชจ๋ธ ๋‹ค์šด๋กœ๋“œ ๋ฐ ์‹คํ–‰\nollama pull qwen2.5:14b\nollama run qwen2.5:14b\n# ์„ค์น˜๋œ ๋ชจ๋ธ ๋ชฉ๋ก\nollama list",
    "common_mistake": "๋ชจ๋ธ ์ด๋ฆ„์— ํƒœ๊ทธ(:14b, :latest ๋“ฑ)๋ฅผ ์ƒ๋žตํ•˜๋ฉด `:latest`๋กœ ์ž๋™ ์„ค์ •๋ฉ๋‹ˆ๋‹ค. ๋ชจ๋ธ์ด ๋กœ์ปฌ์— ์—†์„ ๋•Œ `ollama run`์ด ์ž๋™์œผ๋กœ pullํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["ollama", "pull", "run", "serve", "model tag", "GGUF", "local LLM"],
    "source_id": "official-docs-ollama",
    "details": [
      "**`ollama run`**: ๋ชจ๋ธ์ด ์—†์œผ๋ฉด ์ž๋™ ๋‹ค์šด๋กœ๋“œ ํ›„ ๋Œ€ํ™”ํ˜• ์„ธ์…˜ ์‹œ์ž‘. `--nowordwrap` ์˜ต์…˜์œผ๋กœ ๊ธด ์ถœ๋ ฅ ์ฒ˜๋ฆฌ ๊ฐ€๋Šฅ.",
      "**`ollama serve`**: 11434 ํฌํŠธ์—์„œ OpenAI ํ˜ธํ™˜ REST API ์ œ๊ณต. `OLLAMA_HOST=0.0.0.0`์œผ๋กœ ์™ธ๋ถ€ ์ ‘๊ทผ ํ—ˆ์šฉ ๊ฐ€๋Šฅ.",
      "๋ชจ๋ธ ํŒŒ์ผ์€ `~/.ollama/models`์— GGUF ํ˜•์‹์œผ๋กœ ์ €์žฅ๋ฉ๋‹ˆ๋‹ค. `Modelfile`๋กœ ์ปค์Šคํ…€ ์‹œ์Šคํ…œ ํ”„๋กฌํ”„ํŠธ์™€ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ์„ค์ •ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-ollama-api",
    "certification": "Docs Study",
    "title": "Ollama REST API ํ˜ธ์ถœ",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-3"],
    "summary": "Ollama๋Š” `localhost:11434`์—์„œ REST API๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. `/api/generate`(์™„์„ฑํ˜•), `/api/chat`(๋Œ€ํ™”ํ˜•), `/api/embeddings`(์ž„๋ฒ ๋”ฉ) ์—”๋“œํฌ์ธํŠธ๋ฅผ OpenAI ํ˜ธํ™˜ ํ˜•์‹์œผ๋กœ๋„ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "# chat ์—”๋“œํฌ์ธํŠธ\ncurl http://localhost:11434/api/chat -d '{\n  \"model\": \"qwen2.5:7b\",\n  \"messages\": [{\"role\": \"user\", \"content\": \"์•ˆ๋…•\"}],\n  \"stream\": false\n}'",
    "common_mistake": "Ollama ์„œ๋ฒ„(`ollama serve`)๊ฐ€ ์‹คํ–‰ ์ค‘์ด์ง€ ์•Š๊ฑฐ๋‚˜ ๋ชจ๋ธ์ด ๋กœ์ปฌ์— ์—†์œผ๋ฉด API ํ˜ธ์ถœ์ด ์‹คํŒจํ•ฉ๋‹ˆ๋‹ค. ๋จผ์ € `ollama list`๋กœ ๋ชจ๋ธ ์กด์žฌ๋ฅผ ํ™•์ธํ•˜์„ธ์š”.",
    "keywords": ["ollama", "REST API", "generate", "chat", "embeddings", "localhost:11434", "stream"],
    "source_id": "official-docs-ollama",
    "details": [
      "**OpenAI ํ˜ธํ™˜ API**: `http://localhost:11434/v1/chat/completions`๋กœ OpenAI Python SDK๋‚˜ LangChain์˜ `ChatOpenAI(base_url=...)` ๊ทธ๋Œ€๋กœ ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค.",
      "**์ŠคํŠธ๋ฆฌ๋ฐ**: `\"stream\": true` ์„ค์ • ์‹œ ์‘๋‹ต์„ JSON ์ค„ ๋‹จ์œ„๋กœ ์ŠคํŠธ๋ฆฌ๋ฐ. `\"done\": true` ์ค„์ด ์™„๋ฃŒ ์‹ ํ˜ธ์ž…๋‹ˆ๋‹ค.",
      "**์ž„๋ฒ ๋”ฉ**: `POST /api/embeddings`์— `{\"model\":\"nomic-embed-text\", \"prompt\":\"ํ…์ŠคํŠธ\"}`๋ฅผ ๋ณด๋‚ด๋ฉด ๋ฒกํ„ฐ ๋ฐฐ์—ด์„ ๋ฐ˜ํ™˜ํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-dotenv",
    "certification": "Docs Study",
    "title": "python-dotenv โ€” ํ™˜๊ฒฝ๋ณ€์ˆ˜ & .env ๊ด€๋ฆฌ",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-4"],
    "summary": "python-dotenv๋Š” `.env` ํŒŒ์ผ์˜ ํ‚ค-๊ฐ’ ์Œ์„ `os.environ`์— ์ž๋™์œผ๋กœ ๋กœ๋“œํ•ฉ๋‹ˆ๋‹ค. API ํ‚ค ๊ฐ™์€ ๋ฏผ๊ฐ ์ •๋ณด๋ฅผ ์ฝ”๋“œ์— ์ง์ ‘ ์“ฐ์ง€ ์•Š๊ณ  ํŒŒ์ผ๋กœ ๋ถ„๋ฆฌํ•ด ๊ด€๋ฆฌํ•˜๋Š” ํ‘œ์ค€ ๋ฐฉ๋ฒ•์ž…๋‹ˆ๋‹ค.",
    "example": "# .env ํŒŒ์ผ\nOPENAI_API_KEY=sk-...\nOLLAMA_BASE_URL=http://localhost:11434\n\n# Python ์ฝ”๋“œ\nfrom dotenv import load_dotenv\nimport os\n\nload_dotenv()  # .env ํŒŒ์ผ ๋กœ๋“œ\napi_key = os.getenv('OPENAI_API_KEY')",
    "common_mistake": "`load_dotenv()`๋ฅผ ํ˜ธ์ถœํ•˜๊ธฐ ์ „์— `os.getenv()`๋ฅผ ์“ฐ๋ฉด ํ•ญ์ƒ `None`์„ ๋ฐ˜ํ™˜ํ•ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋“œ์‹œ ํŒŒ์ผ ์ƒ๋‹จ์—์„œ ๊ฐ€์žฅ ๋จผ์ € `load_dotenv()`๋ฅผ ํ˜ธ์ถœํ•˜์„ธ์š”.",
    "keywords": ["dotenv", "load_dotenv", ".env", "os.getenv", "ํ™˜๊ฒฝ๋ณ€์ˆ˜", "API ํ‚ค"],
    "source_id": "official-docs-python",
    "details": [
      "**์šฐ์„ ์ˆœ์œ„**: `load_dotenv(override=False)`(๊ธฐ๋ณธ๊ฐ’)๋Š” ์ด๋ฏธ ์„ค์ •๋œ ํ™˜๊ฒฝ๋ณ€์ˆ˜๋ฅผ ๋ฎ์–ด์“ฐ์ง€ ์•Š์Šต๋‹ˆ๋‹ค. CI/CD ํ™˜๊ฒฝ์—์„œ ์‹œ์Šคํ…œ ํ™˜๊ฒฝ๋ณ€์ˆ˜๊ฐ€ .env๋ณด๋‹ค ์šฐ์„ ํ•ฉ๋‹ˆ๋‹ค.",
      "**.gitignore**: `.env`๋Š” ๋ฐ˜๋“œ์‹œ `.gitignore`์— ์ถ”๊ฐ€ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. `.env.example`์€ ํ‚ค ์ด๋ฆ„๋งŒ ์ ์–ด ํŒ€์›์—๊ฒŒ ์–ด๋–ค ๋ณ€์ˆ˜๊ฐ€ ํ•„์š”ํ•œ์ง€ ์•Œ๋ ค์ฃผ๋Š” ์šฉ๋„๋กœ ์ปค๋ฐ‹ํ•ฉ๋‹ˆ๋‹ค.",
      "**Streamlit ์—ฐ๋™**: ๋กœ์ปฌ ๊ฐœ๋ฐœ ์‹œ `load_dotenv()`๋กœ ํ‚ค๋ฅผ ๋กœ๋“œํ•˜๊ณ , ๋ฐฐํฌ ์‹œ์—๋Š” `st.secrets`๋กœ ์ ‘๊ทผํ•˜๋„๋ก ๋ถ„๊ธฐํ•˜๊ฑฐ๋‚˜ ์ฒ˜์Œ๋ถ€ํ„ฐ `st.secrets`๋งŒ ์“ฐ๋Š” ๋ฐฉ์‹ ์ค‘ ํ•˜๋‚˜๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-prompt-template",
    "certification": "Docs Study",
    "title": "ChatPromptTemplate โ€” ํ”„๋กฌํ”„ํŠธ ๊ตฌ์กฐํ™”",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-5", "docs-lab-6"],
    "summary": "ChatPromptTemplate์€ ์‹œ์Šคํ…œ ํ”„๋กฌํ”„ํŠธ, ์‚ฌ์šฉ์ž ๋ฉ”์‹œ์ง€, ๋ณ€์ˆ˜ ํ”Œ๋ ˆ์ด์Šคํ™€๋”๋ฅผ ๊ตฌ์กฐํ™”๋œ ๋ฐฉ์‹์œผ๋กœ ์ •์˜ํ•ฉ๋‹ˆ๋‹ค. ๋‹จ์ˆœ f-string ๋Œ€์‹  ์žฌ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•˜๊ณ  LangChain ํŒŒ์ดํ”„๋ผ์ธ์— ์—ฐ๊ฒฐ๋˜๋Š” ํ”„๋กฌํ”„ํŠธ๋ฅผ ๋งŒ๋“ค ๋•Œ ์”๋‹ˆ๋‹ค.",
    "example": "from langchain_core.prompts import ChatPromptTemplate\n\nprompt = ChatPromptTemplate.from_messages([\n    (\"system\", \"๋‹น์‹ ์€ ๋„์›€์ด ๋˜๋Š” AI ์–ด์‹œ์Šคํ„ดํŠธ์ž…๋‹ˆ๋‹ค. ์ปจํ…์ŠคํŠธ: {context}\"),\n    (\"human\", \"{question}\")\n])\n\n# ์ฒด์ธ์— ์—ฐ๊ฒฐ\nchain = prompt | llm | StrOutputParser()\nchain.invoke({\"context\": \"...\", \"question\": \"์š”์•ฝํ•ด์ค˜\"})",
    "common_mistake": "`PromptTemplate`์€ ๋‹จ์ˆœ ๋ฌธ์ž์—ด ํฌ๋งท์šฉ(๋ณ€์ˆ˜ ํ•˜๋‚˜), `ChatPromptTemplate`์€ ์—ญํ• ์ด ๊ตฌ๋ถ„๋œ ๋ฉ”์‹œ์ง€ ๋ฐฐ์—ด์šฉ์ž…๋‹ˆ๋‹ค. RAG์—์„œ ์‹œ์Šคํ…œ ํ”„๋กฌํ”„ํŠธ์™€ ์งˆ๋ฌธ์„ ๊ตฌ๋ถ„ํ•˜๋ ค๋ฉด ํ•ญ์ƒ ChatPromptTemplate์„ ์”๋‹ˆ๋‹ค.",
    "keywords": ["ChatPromptTemplate", "from_messages", "system", "human", "placeholder", "PromptTemplate"],
    "source_id": "official-docs-langchain",
    "details": [
      "**MessagesPlaceholder**: ๋Œ€ํ™” ํžˆ์Šคํ† ๋ฆฌ๋ฅผ ํ†ต์งธ๋กœ ์ฃผ์ž…ํ•  ๋•Œ ์”๋‹ˆ๋‹ค. `(\"placeholder\", \"{chat_history}\")`๋กœ ์„ ์–ธํ•˜๋ฉด ๋ฉ”์‹œ์ง€ ๋ฆฌ์ŠคํŠธ๊ฐ€ ๊ทธ ์ž๋ฆฌ์— ํŽผ์ณ์ง‘๋‹ˆ๋‹ค.",
      "**partial**: `prompt.partial(context=\"๊ณ ์ •๊ฐ’\")`์œผ๋กœ ํŠน์ • ๋ณ€์ˆ˜๋ฅผ ๋ฏธ๋ฆฌ ์ฑ„์šธ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. context๋Š” ๊ณ ์ •ํ•˜๊ณ  question๋งŒ ๋Ÿฐํƒ€์ž„์— ๋ฐ›์„ ๋•Œ ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค.",
      "**ํƒ€์ž… ์•ˆ์ „์„ฑ**: from_messages์— `(\"system\", \"...\")` ํŠœํ”Œ ๋Œ€์‹  `SystemMessage(content=\"...\")`๋ฅผ ์‚ฌ์šฉํ•ด๋„ ๋™์ผํ•ฉ๋‹ˆ๋‹ค. ํŠœํ”Œ ๋ฐฉ์‹์ด ๋” ๊ฐ„๊ฒฐํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-document-loaders",
    "certification": "Docs Study",
    "title": "LangChain Document Loaders โ€” ๋ฌธ์„œ ์ผ๊ด„ ๋กœ๋“œ",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-7"],
    "summary": "Document Loader๋Š” PDF, ๋งˆํฌ๋‹ค์šด, ์›น ํŽ˜์ด์ง€, ๋””๋ ‰ํ† ๋ฆฌ ๋“ฑ ๋‹ค์–‘ํ•œ ์†Œ์Šค์—์„œ ๋ฌธ์„œ๋ฅผ LangChain `Document` ๊ฐ์ฒด๋กœ ๋ณ€ํ™˜ํ•ฉ๋‹ˆ๋‹ค. RAG ํŒŒ์ดํ”„๋ผ์ธ์˜ ์ฒซ ๋‹จ๊ณ„์ž…๋‹ˆ๋‹ค.",
    "example": "from langchain_community.document_loaders import DirectoryLoader, PyPDFLoader\n\n# ๋””๋ ‰ํ† ๋ฆฌ ๋‚ด ๋ชจ๋“  .md ํŒŒ์ผ ๋กœ๋“œ\nloader = DirectoryLoader('./notes', glob='**/*.md')\ndocs = loader.load()\n\n# ๊ฐ ๋ฌธ์„œ์˜ ์ถœ์ฒ˜ ํ™•์ธ\nfor doc in docs:\n    print(doc.metadata['source'], len(doc.page_content))",
    "common_mistake": "`loader.load()`๋Š” ๋ชจ๋“  ๋ฌธ์„œ๋ฅผ ๋ฉ”๋ชจ๋ฆฌ์— ์˜ฌ๋ฆฝ๋‹ˆ๋‹ค. ๋Œ€์šฉ๋Ÿ‰ ๋””๋ ‰ํ† ๋ฆฌ์—์„œ๋Š” `loader.lazy_load()`๋กœ ์ œ๋„ˆ๋ ˆ์ดํ„ฐ๋ฅผ ์“ฐ๊ฑฐ๋‚˜ ๋ฐฐ์น˜๋กœ ์ฒ˜๋ฆฌํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["DirectoryLoader", "PyPDFLoader", "WebBaseLoader", "glob", "metadata", "source", "lazy_load"],
    "source_id": "official-docs-langchain",
    "details": [
      "**metadata['source']**: ๋กœ๋“œ๋œ ๋ฌธ์„œ๋Š” `metadata` ๋”•์…”๋„ˆ๋ฆฌ์— ํŒŒ์ผ ๊ฒฝ๋กœ(`source`)๋ฅผ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. RAG ๋‹ต๋ณ€์— ์ถœ์ฒ˜๋ฅผ ํ‘œ์‹œํ•  ๋•Œ ์ด ๊ฐ’์„ ์”๋‹ˆ๋‹ค.",
      "**glob ํŒจํ„ด**: `glob='**/*.md'`๋Š” ํ•˜์œ„ ํด๋”๊นŒ์ง€ ์žฌ๊ท€ ๊ฒ€์ƒ‰ํ•ฉ๋‹ˆ๋‹ค. `glob='*.pdf'`๋Š” ํ˜„์žฌ ํด๋”๋งŒ ๊ฒ€์ƒ‰ํ•ฉ๋‹ˆ๋‹ค.",
      "**WebBaseLoader**: URL ๋ฆฌ์ŠคํŠธ๋ฅผ ๋„˜๊ธฐ๋ฉด ์›น ํŽ˜์ด์ง€๋ฅผ ํฌ๋กค๋งํ•ด Document๋กœ ๋ณ€ํ™˜ํ•ฉ๋‹ˆ๋‹ค. BeautifulSoup ๊ธฐ๋ฐ˜์ด๋ฉฐ `bs_kwargs`๋กœ ํŒŒ์‹ฑ ๋ฒ”์œ„๋ฅผ ์ง€์ •ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-langchain-rag",
    "certification": "Docs Study",
    "title": "LangChain RAG ๊ธฐ๋ณธ ํ๋ฆ„",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-8", "docs-lab-9"],
    "summary": "LangChain์€ ๋ฌธ์„œ ๋กœ๋“œ โ†’ ๋ถ„ํ•  โ†’ ์ž„๋ฒ ๋”ฉ โ†’ ๊ฒ€์ƒ‰ โ†’ ์ƒ์„ฑ์˜ RAG ํŒŒ์ดํ”„๋ผ์ธ์„ ์ปดํฌ๋„ŒํŠธ ์กฐํ•ฉ์œผ๋กœ ๊ตฌํ˜„ํ•ฉ๋‹ˆ๋‹ค. Retriever๊ฐ€ ๊ด€๋ จ ๋ฌธ์„œ๋ฅผ ์ฐพ์•„ LLM ์ปจํ…์ŠคํŠธ๋กœ ์ฃผ์ž…ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from langchain_community.vectorstores import Chroma\nretriever = Chroma.from_documents(docs, embeddings).as_retriever(k=3)\nchain = (\n    {\"context\": retriever, \"question\": RunnablePassthrough()}\n    | prompt | llm | StrOutputParser()\n)",
    "common_mistake": "Retriever๊ฐ€ ์ฐพ์€ ๋ฌธ์„œ๋ฅผ ๊ทธ๋Œ€๋กœ ์ •๋‹ต์œผ๋กœ ๋ฏฟ์ง€ ๋งˆ์„ธ์š”. ๊ฒ€์ƒ‰๋œ ์ฒญํฌ๊ฐ€ ์งˆ๋ฌธ๊ณผ ๋‹ค๋ฅธ ๋งฅ๋ฝ์ผ ์ˆ˜ ์žˆ์–ด ํ”„๋กฌํ”„ํŠธ์—์„œ ๋ช…์‹œ์ ์œผ๋กœ '์•„๋ž˜ ๋ฌธ์„œ๋ฅผ ์ฐธ๊ณ ํ•ด'๋ผ๊ณ  ์ง€์‹œํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["langchain", "RAG", "retriever", "vectorstore", "document loader", "text splitter"],
    "source_id": "official-docs-langchain",
    "details": [
      "**ํŒŒ์ดํ”„๋ผ์ธ ๋‹จ๊ณ„**: DocumentLoader โ†’ TextSplitter(์ฒญํฌ ๋ถ„ํ• ) โ†’ Embeddings โ†’ VectorStore โ†’ Retriever โ†’ LLM Chain.",
      "**TextSplitter**: `RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)`์ด ์ผ๋ฐ˜์  ์‹œ์ž‘์ . overlap์ด ์žˆ์–ด์•ผ ์ฒญํฌ ๊ฒฝ๊ณ„์—์„œ ๋ฌธ๋งฅ ์†์‹ค์„ ์ค„์ž…๋‹ˆ๋‹ค.",
      "`MultiQueryRetriever`๋‚˜ `ContextualCompressionRetriever`๋กœ ๊ฒ€์ƒ‰ ํ’ˆ์งˆ์„ ๋†’์ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹จ์ˆœ ์œ ์‚ฌ๋„ ๊ฒ€์ƒ‰๋ณด๋‹ค ๊ด€๋ จ ์ฒญํฌ ์„ ๋ณ„ ์ •ํ™•๋„๊ฐ€ ์˜ฌ๋ผ๊ฐ‘๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-lcel",
    "certification": "Docs Study",
    "title": "LangChain LCEL โ€” ์ฒด์ธ ๊ตฌ์„ฑ",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-10", "docs-lab-11"],
    "summary": "LCEL(LangChain Expression Language)์€ `|` ํŒŒ์ดํ”„ ์—ฐ์‚ฐ์ž๋กœ Runnable ์ปดํฌ๋„ŒํŠธ๋ฅผ ์ง๋ ฌยท๋ณ‘๋ ฌ ์—ฐ๊ฒฐํ•ด ์ฒด์ธ์„ ์„ ์–ธ์ ์œผ๋กœ ์ •์˜ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from langchain_core.runnables import RunnablePassthrough\nchain = (\n    RunnablePassthrough.assign(context=retriever)\n    | prompt\n    | llm\n    | StrOutputParser()\n)\nresult = chain.invoke({\"question\": \"What is RAG?\"})",
    "common_mistake": "์ฒด์ธ ๊ฐ ๋‹จ๊ณ„์˜ ์ž…์ถœ๋ ฅ ํƒ€์ž…์ด ๋งž์•„์•ผ ํ•ฉ๋‹ˆ๋‹ค. LLM์€ `str` ๋˜๋Š” `BaseMessage`๋ฅผ ๋ฐ›๊ณ  ๋ฐ˜ํ™˜ํ•˜๋ฏ€๋กœ, ์•ž ๋‹จ๊ณ„๊ฐ€ ๋”•์…”๋„ˆ๋ฆฌ๋ฅผ ๋ฐ˜ํ™˜ํ•˜๋ฉด ํ”„๋กฌํ”„ํŠธ ํ…œํ”Œ๋ฆฟ์ด ์ค‘๊ฐ„์—์„œ ๋ฐ›์•„ ์ฒ˜๋ฆฌํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["LCEL", "Runnable", "pipe", "chain", "invoke", "stream", "batch"],
    "source_id": "official-docs-langchain",
    "details": [
      "**3๊ฐ€์ง€ ์‹คํ–‰ ๋ฐฉ์‹**: `invoke`(๋‹จ์ผ ๋™๊ธฐ), `stream`(์ŠคํŠธ๋ฆฌ๋ฐ yield), `batch`(๋ฆฌ์ŠคํŠธ ๋ณ‘๋ ฌ ์ฒ˜๋ฆฌ). ์ŠคํŠธ๋ฆฌ๋ฐ UI์—๋Š” `stream`์„ ์”๋‹ˆ๋‹ค.",
      "**RunnablePassthrough**: ์ž…๋ ฅ์„ ๊ทธ๋Œ€๋กœ ๋‹ค์Œ ๋‹จ๊ณ„๋กœ ์ „๋‹ฌ. `.assign(key=fn)`์œผ๋กœ ์ถ”๊ฐ€ ํ‚ค๋ฅผ ๋ถ™์—ฌ ๋”•์…”๋„ˆ๋ฆฌ๋ฅผ ํ™•์žฅํ•ฉ๋‹ˆ๋‹ค.",
      "**RunnableParallel**: `{\"a\": fn1, \"b\": fn2}` ํ˜•ํƒœ๋กœ ์—ฌ๋Ÿฌ ์ฒด์ธ์„ ๋™์‹œ์— ์‹คํ–‰ํ•ด ๊ฒฐ๊ณผ๋ฅผ ๋”•์…”๋„ˆ๋ฆฌ๋กœ ํ•ฉ์นฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-vectorstore",
    "certification": "Docs Study",
    "title": "VectorStore์™€ Retriever โ€” ๋ฒกํ„ฐ ๊ฒ€์ƒ‰",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-11", "docs-lab-12"],
    "summary": "๋ฌธ์„œ๋ฅผ ์ž„๋ฒ ๋”ฉ ๋ฒกํ„ฐ๋กœ ๋ณ€ํ™˜ํ•ด ์ €์žฅํ•˜๊ณ , ์ฟผ๋ฆฌ์™€ ์ฝ”์‚ฌ์ธ ์œ ์‚ฌ๋„๋กœ ๊ด€๋ จ ์ฒญํฌ๋ฅผ ๊ฒ€์ƒ‰ํ•ฉ๋‹ˆ๋‹ค. Chroma, FAISS, pgvector ๋“ฑ ๋‹ค์–‘ํ•œ ๋ฐฑ์—”๋“œ๋ฅผ LangChain์ด ์ถ”์ƒํ™”ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from langchain_community.vectorstores import Chroma\nfrom langchain_ollama import OllamaEmbeddings\n\nembeddings = OllamaEmbeddings(model='nomic-embed-text')\ndb = Chroma.from_documents(docs, embeddings, persist_directory='./db')\nretriever = db.as_retriever(search_type='mmr', search_kwargs={'k': 4})",
    "common_mistake": "์ž„๋ฒ ๋”ฉ ๋ชจ๋ธ์„ ๋ณ€๊ฒฝํ•˜๋ฉด ๋ฒกํ„ฐ ์ฐจ์›์ด ๋‹ฌ๋ผ์ ธ ๊ธฐ์กด ์ปฌ๋ ‰์…˜๊ณผ ํ˜ธํ™˜๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋ชจ๋ธ์„ ๋ฐ”๊ฟ€ ๋•Œ๋Š” ์ปฌ๋ ‰์…˜์„ ์ƒˆ๋กœ ๋งŒ๋“ค์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["vectorstore", "Chroma", "FAISS", "embedding", "similarity", "MMR", "retriever"],
    "source_id": "official-docs-langchain",
    "details": [
      "**๊ฒ€์ƒ‰ ์œ ํ˜•**: `similarity`(์ˆœ์ˆ˜ ์œ ์‚ฌ๋„ ์ƒ์œ„ k๊ฐœ), `mmr`(MMR: ๋‹ค์–‘์„ฑ ๊ณ ๋ ค), `similarity_score_threshold`(์ž„๊ณ„๊ฐ’ ์ด์ƒ๋งŒ).",
      "**Chroma**: ๋กœ์ปฌ ํŒŒ์ผ๋กœ ์ €์žฅ ๊ฐ€๋Šฅ(`persist_directory`). ํ”„๋กœํ† ํƒ€์ž…์— ์ ํ•ฉ. ํ”„๋กœ๋•์…˜์€ pgvector(PostgreSQL), Qdrant, Pinecone ๊ถŒ์žฅ.",
      "์ž„๋ฒ ๋”ฉ์€ ํ…์ŠคํŠธ์˜ ์˜๋ฏธ์  ์œ ์‚ฌ๋„๋ฅผ ์žก์ง€๋งŒ, ์ •ํ™•ํ•œ ํ‚ค์›Œ๋“œ ๋งค์นญ์ด ํ•„์š”ํ•˜๋ฉด BM25 ๋“ฑ ํ‚ค์›Œ๋“œ ๊ฒ€์ƒ‰๊ณผ `EnsembleRetriever`๋กœ ํ•˜์ด๋ธŒ๋ฆฌ๋“œ ๊ตฌ์„ฑ์„ ๊ณ ๋ คํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-streamlit-state",
    "certification": "Docs Study",
    "title": "Streamlit session_state โ€” ์ƒํƒœ ๊ด€๋ฆฌ",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-16", "docs-lab-17"],
    "summary": "Streamlit์€ ์ƒํ˜ธ์ž‘์šฉ๋งˆ๋‹ค ์Šคํฌ๋ฆฝํŠธ ์ „์ฒด๋ฅผ rerunํ•ฉ๋‹ˆ๋‹ค. ๋ฒ„ํŠผ ํด๋ฆญยท์ž…๋ ฅ ๊ฒฐ๊ณผ๋ฅผ rerun ์ดํ›„์—๋„ ์œ ์ง€ํ•˜๋ ค๋ฉด `st.session_state`์— ์ €์žฅํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "if 'count' not in st.session_state:\n    st.session_state.count = 0\nif st.button('์ฆ๊ฐ€'):\n    st.session_state.count += 1\nst.write(f'ํด๋ฆญ ํšŸ์ˆ˜: {st.session_state.count}')",
    "common_mistake": "์ผ๋ฐ˜ ์ „์—ญ ๋ณ€์ˆ˜๋‚˜ ํ•จ์ˆ˜ ๋‚ด ์ง€์—ญ ๋ณ€์ˆ˜๋Š” rerun๋งˆ๋‹ค ์ดˆ๊ธฐํ™”๋ฉ๋‹ˆ๋‹ค. ์ƒํƒœ ๋ณด์กด์ด ํ•„์š”ํ•œ ๋ชจ๋“  ๊ฐ’์€ ๋ฐ˜๋“œ์‹œ `st.session_state`๋ฅผ ์‚ฌ์šฉํ•˜์„ธ์š”.",
    "keywords": ["session_state", "rerun", "widget", "state management", "callback"],
    "source_id": "official-docs-streamlit",
    "details": [
      "**์ดˆ๊ธฐํ™” ํŒจํ„ด**: `st.session_state.setdefault('key', default_value)` ๋˜๋Š” `if 'key' not in st.session_state:` ๋ธ”๋ก์œผ๋กœ ์•ˆ์ „ํ•˜๊ฒŒ ์ดˆ๊ธฐํ™”ํ•ฉ๋‹ˆ๋‹ค.",
      "**Widget key**: `st.text_input('์ด๋ฆ„', key='username')`์œผ๋กœ ์œ„์ ฏ ๊ฐ’์ด ์ž๋™์œผ๋กœ `st.session_state.username`์— ์ €์žฅ๋ฉ๋‹ˆ๋‹ค.",
      "**callback**: `on_change=my_func`์œผ๋กœ ์œ„์ ฏ ๋ณ€๊ฒฝ ์‹œ ํ•จ์ˆ˜๋ฅผ ์‹คํ–‰ํ•ฉ๋‹ˆ๋‹ค. callback ์•ˆ์—์„œ session_state๋ฅผ ์ˆ˜์ •ํ•˜๋ฉด rerun ์ „์— ์ฒ˜๋ฆฌ๋ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-streamlit-cache",
    "certification": "Docs Study",
    "title": "Streamlit Cache โ€” ์บ์‹œ ๋ฐ์ฝ”๋ ˆ์ดํ„ฐ",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-18"],
    "summary": "rerun๋งˆ๋‹ค ๋ฐ˜๋ณต ์‹คํ–‰๋˜๋Š” ๋น„์šฉ ๋†’์€ ๊ณ„์‚ฐยท๋ฆฌ์†Œ์Šค ๋กœ๋”ฉ์„ ์บ์‹œํ•ด ์„ฑ๋Šฅ์„ ๋†’์ž…๋‹ˆ๋‹ค. `@st.cache_data`๋Š” ์ง๋ ฌํ™” ๊ฐ€๋Šฅ ๋ฐ์ดํ„ฐ, `@st.cache_resource`๋Š” DBยท๋ชจ๋ธ ๊ฐ™์€ ๊ณต์œ  ๊ฐ์ฒด์— ์”๋‹ˆ๋‹ค.",
    "example": "@st.cache_resource\ndef load_model():\n    return SentenceTransformer('all-MiniLM-L6-v2')\n\n@st.cache_data(ttl=300)\ndef fetch_data(query: str):\n    return db.execute(query).fetchall()",
    "common_mistake": "`cache_data`์™€ `cache_resource`๋ฅผ ํ˜ผ๋™ํ•˜๋ฉด ๋ฌธ์ œ๊ฐ€ ์ƒ๊น๋‹ˆ๋‹ค. LLM ๋ชจ๋ธ์ด๋‚˜ DB ์—ฐ๊ฒฐ์ฒ˜๋Ÿผ ๊ณต์œ ํ•ด์•ผ ํ•˜๋Š” ๊ฐ์ฒด๋Š” ๋ฐ˜๋“œ์‹œ `cache_resource`๋ฅผ ์จ์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["cache_data", "cache_resource", "TTL", "rerun", "performance"],
    "source_id": "official-docs-streamlit",
    "details": [
      "**`@st.cache_data`**: ํ•จ์ˆ˜ ์ธ์ˆ˜๋ฅผ ํ‚ค๋กœ ๊ฒฐ๊ณผ๋ฅผ ๋ณต์‚ฌํ•ด ์บ์‹œ. DataFrame, ๋ฆฌ์ŠคํŠธ ๋“ฑ ์ง๋ ฌํ™” ๊ฐ€๋Šฅ ๋ฐ˜ํ™˜๊ฐ’์— ์ ํ•ฉ. `ttl` ์ธ์ˆ˜๋กœ ๋งŒ๋ฃŒ ์‹œ๊ฐ„ ์„ค์ • ๊ฐ€๋Šฅ.",
      "**`@st.cache_resource`**: ์‹ฑ๊ธ€ํ†ค์ฒ˜๋Ÿผ ํ•œ ๋ฒˆ๋งŒ ์ƒ์„ฑํ•ด ๋ชจ๋“  ์„ธ์…˜์ด ๊ณต์œ . ๋ชจ๋ธ ๋กœ๋”ฉ, DB ์—ฐ๊ฒฐ, ๋ฒกํ„ฐ ์Šคํ† ์–ด ์ดˆ๊ธฐํ™”์— ์ ํ•ฉ.",
      "์บ์‹œ ๋ฌดํšจํ™”: `st.cache_data.clear()` / `st.cache_resource.clear()`๋กœ ์ˆ˜๋™ ์ดˆ๊ธฐํ™”. ํ•จ์ˆ˜ ๋ณธ๋ฌธ์ด ๋ฐ”๋€Œ๋ฉด ์ž๋™์œผ๋กœ ์บ์‹œ๊ฐ€ ๋ฌดํšจํ™”๋ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-streamlit-secrets",
    "certification": "Docs Study",
    "title": "Streamlit Secrets โ€” ๋น„๋ฐ€ ๊ด€๋ฆฌ",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-4"],
    "summary": "API ํ‚ค, DB ๋น„๋ฐ€๋ฒˆํ˜ธ ๋“ฑ ๋ฏผ๊ฐ ์ •๋ณด๋ฅผ ์ฝ”๋“œ์— ๋„ฃ์ง€ ์•Š๊ณ  `.streamlit/secrets.toml` ํŒŒ์ผ ๋˜๋Š” Streamlit Cloud์˜ Secrets UI์— ์ €์žฅํ•ฉ๋‹ˆ๋‹ค.",
    "example": "# .streamlit/secrets.toml\n[database]\nhost = 'localhost'\npassword = 'secret'\n\n# ์ฝ”๋“œ์—์„œ ์ ‘๊ทผ\nimport streamlit as st\ndb_pass = st.secrets['database']['password']",
    "common_mistake": "`.streamlit/secrets.toml`์„ `.gitignore`์— ์ถ”๊ฐ€ํ•˜์ง€ ์•Š์œผ๋ฉด ๋ฏผ๊ฐ ์ •๋ณด๊ฐ€ Git์— ์ปค๋ฐ‹๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฐ˜๋“œ์‹œ ์ œ์™ธํ•˜์„ธ์š”.",
    "keywords": ["secrets", "secrets.toml", "API key", ".gitignore", "Streamlit Cloud"],
    "source_id": "official-docs-streamlit",
    "details": [
      "๋กœ์ปฌ ๊ฐœ๋ฐœ: `.streamlit/secrets.toml`์— TOML ํ˜•์‹์œผ๋กœ ์ €์žฅ. ์•ฑ ์ฝ”๋“œ์—์„œ `st.secrets`๋กœ ๋”•์…”๋„ˆ๋ฆฌ์ฒ˜๋Ÿผ ์ ‘๊ทผํ•ฉ๋‹ˆ๋‹ค.",
      "Streamlit Cloud: ์•ฑ Settings โ†’ Secrets์— ๊ฐ™์€ TOML ๋‚ด์šฉ์„ ๋ถ™์—ฌ๋„ฃ์œผ๋ฉด ๋ฐฐํฌ ์‹œ ์ž๋™ ์ฃผ์ž…๋ฉ๋‹ˆ๋‹ค.",
      "`st.secrets`๋Š” `os.environ`๋ณด๋‹ค ๊ตฌ์กฐํ™”๋œ ์ ‘๊ทผ์„ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค. `st.secrets['key']`์™€ `st.secrets.key` ๋‘ ๋ฐฉ์‹ ๋ชจ๋‘ ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-hf-spaces",
    "certification": "Docs Study",
    "title": "Hugging Face Spaces โ€” ์•ฑ ๋ฐฐํฌ",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-19"],
    "summary": "Hugging Face Spaces๋Š” Streamlit, Gradio, Docker ์•ฑ์„ ๋ฌด๋ฃŒ๋กœ ํ˜ธ์ŠคํŒ…ํ•˜๋Š” ํ™˜๊ฒฝ์ž…๋‹ˆ๋‹ค. README.md ์ƒ๋‹จ์˜ YAML ๋ฉ”ํƒ€๋ฐ์ดํ„ฐ๋กœ SDK์™€ ํŒŒ์ผ์„ ์ง€์ •ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "# README.md ์ƒ๋‹จ (YAML frontmatter)\n---\ntitle: My App\nsdk: streamlit\napp_file: app.py\npython_version: '3.11'\n---",
    "common_mistake": "API ํ‚ค ๋“ฑ ๋ฏผ๊ฐ ์ •๋ณด๋Š” ์ฝ”๋“œ์— ํ•˜๋“œ์ฝ”๋”ฉํ•˜์ง€ ๋งˆ์„ธ์š”. Space Settings โ†’ Secrets์— ๋“ฑ๋กํ•˜๊ณ  ์ฝ”๋“œ์—์„œ๋Š” `os.environ['MY_KEY']`๋กœ ์ฐธ์กฐํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["huggingface", "spaces", "SDK", "secrets", "app_file", "README metadata"],
    "source_id": "official-docs-huggingface",
    "details": [
      "**๋ฌด๋ฃŒ ํ‹ฐ์–ด**: CPU ๊ธฐ๋ณธ ์ œ๊ณต, ๊ณต๊ฐœ Space๋Š” ๋ฌด๋ฃŒ. GPU๋‚˜ ๋น„๊ณต๊ฐœ Space๋Š” ์œ ๋ฃŒ ํ”Œ๋žœ ๋˜๋Š” ZeroGPU(๋ฌด๋ฃŒ ๊ณต์œ  GPU) ํ™œ์šฉ.",
      "**Secrets**: `Settings โ†’ Repository secrets`์— ํ‚ค-๊ฐ’์œผ๋กœ ๋“ฑ๋ก. ๋นŒ๋“œ ๋ฐ ๋Ÿฐํƒ€์ž„์— ํ™˜๊ฒฝ ๋ณ€์ˆ˜๋กœ ์ฃผ์ž…๋ฉ๋‹ˆ๋‹ค.",
      "Space๋ฅผ Duplicate(๋ณต์ œ)ํ•˜๋ฉด fork์ฒ˜๋Ÿผ ๋…๋ฆฝ๋œ Space๊ฐ€ ์ƒ์„ฑ๋ฉ๋‹ˆ๋‹ค. ํƒ€์ธ์˜ ๋ชจ๋ธยท์•ฑ์„ ๋‚ด Secrets์™€ ํ•จ๊ป˜ ์ปค์Šคํ…€ํ•˜๋Š” ์ผ๋ฐ˜์ ์ธ ํŒจํ„ด์ž…๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-hf-docker-space",
    "certification": "Docs Study",
    "title": "Hugging Face Docker Space",
    "level": "์ž…๋ฌธ",
    "related_practices": ["docs-lab-20"],
    "summary": "Dockerfile๋กœ ์•ฑ ํ™˜๊ฒฝ์„ ์ง์ ‘ ์ •์˜ํ•˜๋Š” Space ๋ฐฐํฌ ๋ฐฉ์‹์ž…๋‹ˆ๋‹ค. Streamlit/Gradio SDK ์ œ์•ฝ์„ ๋„˜์–ด FastAPI, ์ปค์Šคํ…€ ์„œ๋ฒ„ ๋“ฑ ์–ด๋–ค ์•ฑ๋„ ๋ฐฐํฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
    "example": "# README.md\n---\ntitle: My API\nsdk: docker\napp_port: 7860\n---\n\n# Dockerfile\nFROM python:3.11-slim\nCOPY . .\nRUN pip install -r requirements.txt\nEXPOSE 7860\nCMD [\"uvicorn\", \"main:app\", \"--host\", \"0.0.0.0\", \"--port\", \"7860\"]",
    "common_mistake": "Docker Space๋Š” ํฌํŠธ 7860์„ ๊ธฐ๋ณธ์œผ๋กœ ๋…ธ์ถœํ•ฉ๋‹ˆ๋‹ค(`app_port: 7860`). ์•ฑ์ด ๋‹ค๋ฅธ ํฌํŠธ๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด README ๋ฉ”ํƒ€๋ฐ์ดํ„ฐ์˜ `app_port`์™€ ์ผ์น˜์‹œ์ผœ์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["docker", "Dockerfile", "Spaces", "app_port", "FastAPI", "custom server"],
    "source_id": "official-docs-huggingface",
    "details": [
      "**๋นŒ๋“œ ์บ์‹œ**: Space๋Š” ์ด์ „ ๋นŒ๋“œ ๋ ˆ์ด์–ด๋ฅผ ์บ์‹œํ•ฉ๋‹ˆ๋‹ค. `COPY requirements.txt .` ํ›„ `RUN pip install`์„ ๋ฐฐ์น˜ํ•ด ์˜์กด์„ฑ ๋ ˆ์ด์–ด๋ฅผ ์•ฑ ์ฝ”๋“œ๋ณด๋‹ค ์•ž์— ๋‘๋ฉด ๋นŒ๋“œ๊ฐ€ ๋น ๋ฆ…๋‹ˆ๋‹ค.",
      "**Secrets**: `docker build --build-arg`๊ฐ€ ์•„๋‹Œ Runtime Secrets๋กœ ์ฃผ์ž…ํ•ด์•ผ ์ด๋ฏธ์ง€์— ๋ฏผ๊ฐ ์ •๋ณด๊ฐ€ ํฌํ•จ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. `os.environ`์œผ๋กœ ์ ‘๊ทผํ•ฉ๋‹ˆ๋‹ค.",
      "**๋ชจ๋ธ ํŒŒ์ผ**: ๋Œ€ํ˜• ๋ชจ๋ธ์€ Space ์ €์žฅ์†Œ์— ์ง์ ‘ ๋„ฃ์ง€ ๋ง๊ณ  `huggingface_hub.snapshot_download()`๋กœ ๋Ÿฐํƒ€์ž„์— ๊ฐ€์ ธ์˜ค๊ฑฐ๋‚˜ Dataset ์ €์žฅ์†Œ๋ฅผ ๋งˆ์šดํŠธํ•˜๋Š” ๋ฐฉ์‹์„ ์”๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-fastapi",
    "certification": "Docs Study",
    "title": "FastAPI โ€” LLM ์„œ๋น„์Šค API ์„œ๋ฒ„",
    "level": "์ค‘๊ธ‰",
    "related_practices": ["docs-lab-21"],
    "summary": "FastAPI๋Š” Python ํƒ€์ž… ํžŒํŠธ ๊ธฐ๋ฐ˜์˜ ๊ณ ์„ฑ๋Šฅ ๋น„๋™๊ธฐ API ํ”„๋ ˆ์ž„์›Œํฌ์ž…๋‹ˆ๋‹ค. RAG ์ฒด์ธ์ด๋‚˜ LLM ๊ธฐ๋Šฅ์„ REST API ์—”๋“œํฌ์ธํŠธ๋กœ ๋…ธ์ถœํ•  ๋•Œ ๊ฐ€์žฅ ๋งŽ์ด ์“ฐ์ž…๋‹ˆ๋‹ค.",
    "example": "from fastapi import FastAPI\nfrom pydantic import BaseModel\n\napp = FastAPI()\n\nclass Query(BaseModel):\n    question: str\n\n@app.post('/ask')\nasync def ask(q: Query):\n    answer = chain.invoke(q.question)\n    return {\"answer\": answer}\n\n# ์‹คํ–‰\n# uvicorn main:app --reload",
    "common_mistake": "LangChain ์ฒด์ธ ๊ฐ™์€ ๋ฌด๊ฑฐ์šด ๊ฐ์ฒด๋ฅผ ์š”์ฒญ๋งˆ๋‹ค ์ƒ์„ฑํ•˜์ง€ ๋งˆ์„ธ์š”. ๋ชจ๋“ˆ ๋ ˆ๋ฒจ์—์„œ ํ•œ ๋ฒˆ ์ดˆ๊ธฐํ™”ํ•˜๊ฑฐ๋‚˜ `lifespan` ์ด๋ฒคํŠธ๋กœ ์•ฑ ์‹œ์ž‘ ์‹œ ๋กœ๋“œํ•˜์„ธ์š”.",
    "keywords": ["fastapi", "uvicorn", "pydantic", "async", "REST API", "endpoint", "lifespan"],
    "source_id": "official-docs-fastapi",
    "details": [
      "**Pydantic**: ์š”์ฒญ/์‘๋‹ต ์Šคํ‚ค๋งˆ๋ฅผ `BaseModel`๋กœ ์ •์˜ํ•˜๋ฉด ์ž๋™ ๊ฒ€์ฆ, OpenAPI ๋ฌธ์„œ(`/docs`), JSON ์ง๋ ฌํ™”๊ฐ€ ๋ฌด๋ฃŒ๋กœ ์ œ๊ณต๋ฉ๋‹ˆ๋‹ค.",
      "**์ŠคํŠธ๋ฆฌ๋ฐ ์‘๋‹ต**: `StreamingResponse`์™€ async generator๋กœ LLM ์ถœ๋ ฅ์„ Server-Sent Events(SSE)๋กœ ์ŠคํŠธ๋ฆฌ๋ฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
      "**lifespan**: `@asynccontextmanager`๋กœ ์•ฑ ์‹œ์ž‘ ์‹œ ๋ชจ๋ธยทDB ์—ฐ๊ฒฐ์„ ์ดˆ๊ธฐํ™”ํ•˜๊ณ  ์ข…๋ฃŒ ์‹œ ์ •๋ฆฌํ•˜๋Š” ํŒจํ„ด. FastAPI 0.95+ ๊ถŒ์žฅ ๋ฐฉ์‹์ž…๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-langsmith",
    "certification": "Docs Study",
    "title": "LangSmith โ€” LLM ์ถ”์  & ํ‰๊ฐ€",
    "level": "์ค‘๊ธ‰",
    "related_practices": ["docs-lab-13"],
    "summary": "LangSmith๋Š” LangChain ํŒŒ์ดํ”„๋ผ์ธ์˜ ๋ชจ๋“  LLM ํ˜ธ์ถœ, ํ”„๋กฌํ”„ํŠธ, ์‘๋‹ต, ์ง€์—ฐ์‹œ๊ฐ„์„ ์ž๋™ ์ถ”์ ํ•ฉ๋‹ˆ๋‹ค. ๋ฒ„๊ทธ ๋””๋ฒ„๊น…, ํ”„๋กฌํ”„ํŠธ ๋น„๊ต, ํ‰๊ฐ€ ๋ฐ์ดํ„ฐ์…‹ ๊ด€๋ฆฌ๋ฅผ ํ•œ ๊ณณ์—์„œ ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "import os\nos.environ['LANGCHAIN_TRACING_V2'] = 'true'\nos.environ['LANGCHAIN_API_KEY'] = 'ls__...'\nos.environ['LANGCHAIN_PROJECT'] = 'my-rag-app'\n\n# ์ดํ›„ LangChain ์ฝ”๋“œ๋Š” ์ž๋™์œผ๋กœ ์ถ”์ ๋จ\nresult = chain.invoke({'question': '์งˆ๋ฌธ'})",
    "common_mistake": "์ถ”์  ํ™œ์„ฑํ™”๋Š” ํ™˜๊ฒฝ๋ณ€์ˆ˜ 3๊ฐœ(`TRACING_V2=true`, `API_KEY`, `PROJECT`)๊ฐ€ ๋ชจ๋‘ ์žˆ์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ํ•˜๋‚˜๋ผ๋„ ๋น ์ง€๋ฉด ์กฐ์šฉํžˆ ์ถ”์ ์ด ์•ˆ ๋ฉ๋‹ˆ๋‹ค.",
    "keywords": ["langsmith", "tracing", "LangChain", "debugging", "evaluation", "dataset", "prompt"],
    "source_id": "official-docs-langchain",
    "details": [
      "**Run ํŠธ๋ ˆ์ด์Šค**: ์ฒด์ธ ์‹คํ–‰๋งˆ๋‹ค tree ํ˜•ํƒœ๋กœ ๊ฐ ๋‹จ๊ณ„์˜ ์ž…์ถœ๋ ฅยทํ† ํฐ ์ˆ˜ยท์ง€์—ฐ์‹œ๊ฐ„์„ ๊ธฐ๋กํ•ฉ๋‹ˆ๋‹ค. ์–ด๋А ๋‹จ๊ณ„์—์„œ ์˜ค๋ฅ˜๊ฐ€ ๋‚ฌ๋Š”์ง€ ์ฆ‰์‹œ ํŒŒ์•…ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
      "**๋ฐ์ดํ„ฐ์…‹ & ํ‰๊ฐ€**: ํ”„๋กœ๋•์…˜ ์˜ค๋ฅ˜ ์ผ€์ด์Šค๋ฅผ Dataset์œผ๋กœ ์ €์žฅ โ†’ Evaluator(LLM-as-judge ๋“ฑ)๋กœ ์ž๋™ ํ‰๊ฐ€ โ†’ CI์— ํ†ตํ•ฉํ•˜๋Š” ํ๋ฆ„์„ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค.",
      "**Playground**: LangSmith UI์—์„œ ํŠน์ • Run์˜ ํ”„๋กฌํ”„ํŠธ๋ฅผ ๋ฐ”๋กœ ์ˆ˜์ •ํ•ด ์žฌ์‹คํ–‰ ๊ฐ€๋Šฅ. ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง ๋ฐ˜๋ณต์— ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-guardrails",
    "certification": "Docs Study",
    "title": "Guardrails AI โ€” LLM ์ถœ๋ ฅ ๊ฒ€์ฆ",
    "level": "์ค‘๊ธ‰",
    "related_practices": ["docs-lab-14"],
    "summary": "Guardrails AI๋Š” LLM ์ž…๋ ฅ/์ถœ๋ ฅ์— ๊ฒ€์ฆ ๊ทœ์น™(Guard)์„ ์ ์šฉํ•ฉ๋‹ˆ๋‹ค. ์œ ํ•ด ์ฝ˜ํ…์ธ  ์ฐจ๋‹จ, JSON ์Šคํ‚ค๋งˆ ๊ฐ•์ œ, PII ๋งˆ์Šคํ‚น ๋“ฑ ํ”„๋กœ๋•์…˜ LLM ์•ฑ์˜ ์•ˆ์ „์žฅ์น˜ ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from guardrails import Guard\nfrom guardrails.hub import ToxicLanguage\n\nguard = Guard().use(ToxicLanguage(threshold=0.5, on_fail='exception'))\n\nresult = guard(\n    llm_api=openai.chat.completions.create,\n    prompt='์‚ฌ์šฉ์ž ์ž…๋ ฅ: {{user_input}}',\n    user_input=user_message\n)",
    "common_mistake": "Guard๋Š” LLM ํ˜ธ์ถœ์„ ๋ž˜ํ•‘ํ•˜๋ฏ€๋กœ ์‘๋‹ต ์‹œ๊ฐ„์ด ์ฆ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ๋ชจ๋“  ์š”์ฒญ์— heavy Guard๋ฅผ ๊ฑธ๋ฉด ์ง€์—ฐ์ด ์ปค์ง€๋‹ˆ, ์ž…๋ ฅ ๊ฒ€์ฆ๊ณผ ์ถœ๋ ฅ ๊ฒ€์ฆ์„ ๋ถ„๋ฆฌํ•ด ํ•„์š”ํ•œ ๊ฒƒ๋งŒ ์ ์šฉํ•˜์„ธ์š”.",
    "keywords": ["guardrails", "guard", "validation", "toxic", "PII", "JSON schema", "on_fail", "safety"],
    "source_id": "official-docs-guardrails",
    "details": [
      "**Guard ํ—ˆ๋ธŒ**: `guardrails hub install hub://...`์œผ๋กœ ์ปค๋ฎค๋‹ˆํ‹ฐ Guard๋ฅผ ์„ค์น˜ํ•ฉ๋‹ˆ๋‹ค. ToxicLanguage, ValidJson, DetectPII, RegexMatch ๋“ฑ ์ˆ˜์‹ญ ๊ฐ€์ง€๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.",
      "**on_fail ๋™์ž‘**: `'exception'`(์˜ˆ์™ธ ๋ฐœ์ƒ), `'reask'`(LLM์— ์žฌ์š”์ฒญ), `'fix'`(์ž๋™ ์ˆ˜์ •), `'noop'`(ํ†ต๊ณผ) ์ค‘ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.",
      "**๊ตฌ์กฐํ™” ์ถœ๋ ฅ**: `Guard.from_pydantic(OutputModel)`์œผ๋กœ LLM์ด ํ•ญ์ƒ ์ง€์ •ํ•œ ์Šคํ‚ค๋งˆ์˜ JSON์„ ๋ฐ˜ํ™˜ํ•˜๋„๋ก ๊ฐ•์ œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-gradio",
    "certification": "Docs Study",
    "title": "Gradio โ€” ML ๋ฐ๋ชจ UI",
    "level": "์ค‘๊ธ‰",
    "related_practices": ["docs-lab-22"],
    "summary": "Gradio๋Š” ML ๋ชจ๋ธ ๋ฐ๋ชจ๋ฅผ ๋ช‡ ์ค„๋กœ ๋งŒ๋“œ๋Š” UI ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ์ž…๋‹ˆ๋‹ค. `gr.Interface`๋กœ ํ•จ์ˆ˜๋ฅผ ๊ฐ์‹ธ๋ฉด ์ฆ‰์‹œ ์›น UI๊ฐ€ ์ƒ๊ธฐ๊ณ , HuggingFace Spaces์™€ ๋„ค์ดํ‹ฐ๋ธŒ ํ†ตํ•ฉ๋ฉ๋‹ˆ๋‹ค.",
    "example": "import gradio as gr\n\ndef chat(message, history):\n    response = chain.invoke(message)\n    return response\n\ndemo = gr.ChatInterface(\n    fn=chat,\n    title='RAG ์ฑ—๋ด‡',\n    examples=['๋ฌธ์„œ ์š”์•ฝํ•ด์ค˜', '์ฃผ์š” ํ‚ค์›Œ๋“œ๋Š”?']\n)\ndemo.launch()",
    "common_mistake": "Streamlit๊ณผ ๋‹ฌ๋ฆฌ Gradio๋Š” ํ•จ์ˆ˜ ๋‹จ์œ„๋กœ UI๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๋ณต์žกํ•œ ์ƒํƒœ ๊ด€๋ฆฌ๋Š” `gr.State()`๋กœ, ์—ฌ๋Ÿฌ ์ปดํฌ๋„ŒํŠธ๋ฅผ ์กฐํ•ฉํ•  ๋•Œ๋Š” `gr.Blocks()` ์ปจํ…์ŠคํŠธ๋ฅผ ์”๋‹ˆ๋‹ค.",
    "keywords": ["gradio", "Interface", "Blocks", "ChatInterface", "State", "HuggingFace", "demo"],
    "source_id": "official-docs-gradio",
    "details": [
      "**Interface vs Blocks**: `gr.Interface`๋Š” ์ž…๋ ฅโ†’ํ•จ์ˆ˜โ†’์ถœ๋ ฅ ๋‹จ์ˆœ ๊ตฌ์กฐ, `gr.Blocks`๋Š” ๋ ˆ์ด์•„์›ƒ๊ณผ ์ด๋ฒคํŠธ๋ฅผ ์ž์œ ๋กญ๊ฒŒ ๊ตฌ์„ฑ. LLM ์ฑ—๋ด‡์—๋Š” `gr.ChatInterface`๊ฐ€ ๊ฐ€์žฅ ์ ํ•ฉํ•ฉ๋‹ˆ๋‹ค.",
      "**์ŠคํŠธ๋ฆฌ๋ฐ**: `gr.ChatInterface(fn=chat)`์—์„œ `chat`์„ generator ํ•จ์ˆ˜๋กœ ๋งŒ๋“ค๊ณ  `yield`๋กœ ํ† ํฐ์„ ๋‚ด๋ณด๋‚ด๋ฉด ์ž๋™์œผ๋กœ ์ŠคํŠธ๋ฆฌ๋ฐ UI๊ฐ€ ๋ฉ๋‹ˆ๋‹ค.",
      "**HF Spaces ํ†ตํ•ฉ**: `demo.launch()`๋งŒ ํ•˜๋ฉด HF Spaces Gradio SDK๋กœ ์ฆ‰์‹œ ๋ฐฐํฌ๋ฉ๋‹ˆ๋‹ค. ๋ณ„๋„ ์„œ๋ฒ„ ์„ค์ • ์—†์ด `sdk: gradio`๋งŒ README์— ์„ ์–ธํ•˜๋ฉด ๋ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-chroma",
    "certification": "Docs Study",
    "title": "Chroma โ€” ์˜์† ๋ฒกํ„ฐ DB",
    "level": "์ค‘๊ธ‰",
    "related_practices": ["docs-chroma-1", "docs-chroma-2"],
    "summary": "Chroma๋Š” ์žฌ์‹œ์ž‘ํ•ด๋„ ๋ฐ์ดํ„ฐ๊ฐ€ ์œ ์ง€๋˜๋Š” ์˜คํ”ˆ์†Œ์Šค ๋ฒกํ„ฐ DB์ž…๋‹ˆ๋‹ค. FAISS๋Š” `save_local()`์„ ์ง์ ‘ ํ˜ธ์ถœํ•ด์•ผ ํ•˜์ง€๋งŒ, Chroma๋Š” `persist_directory`๋งŒ ์ง€์ •ํ•˜๋ฉด ์ž๋™์œผ๋กœ ๋กœ์ปฌ DB์— ์ €์žฅํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from langchain_chroma import Chroma\n\n# ์ƒ์„ฑ & ์ž๋™ ์ €์žฅ\ndb = Chroma.from_documents(\n    chunks,\n    embeddings,\n    persist_directory='./chroma_db'\n)\n\n# ๋‹ค์Œ ์‹คํ–‰ ์‹œ โ€” ์ž„๋ฒ ๋”ฉ ์žฌ๊ณ„์‚ฐ ์—†์ด ๋กœ๋“œ\ndb = Chroma(\n    persist_directory='./chroma_db',\n    embedding_function=embeddings\n)",
    "common_mistake": "FAISS์ฒ˜๋Ÿผ `save_local()`์„ ๋ณ„๋„๋กœ ํ˜ธ์ถœํ•  ํ•„์š”๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค. `persist_directory`๋ฅผ ์ง€์ •ํ•˜๋ฉด ์ž๋™ ์ €์žฅ๋ฉ๋‹ˆ๋‹ค. ๋‹จ, `persist_directory`๋ฅผ ์ƒ๋žตํ•˜๋ฉด ์ธ๋ฉ”๋ชจ๋ฆฌ๋กœ๋งŒ ๋™์ž‘ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["Chroma", "persist_directory", "from_documents", "collection", "chromadb", "์˜์†"],
    "source_id": "official-docs-chromadb",
    "details": [
      "**FAISS vs Chroma**: FAISS๋Š” ์ธ๋ฉ”๋ชจ๋ฆฌ ๊ธฐ๋ฐ˜์œผ๋กœ `save_local()`๋กœ ์ˆ˜๋™ ์ €์žฅํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. Chroma๋Š” `persist_directory`๋งŒ ์ง€์ •ํ•˜๋ฉด SQLite ๊ธฐ๋ฐ˜ ๋กœ์ปฌ DB์— ์ž๋™ ์ €์žฅ๋ฉ๋‹ˆ๋‹ค.",
      "**์ปฌ๋ ‰์…˜**: Chroma๋Š” ๋ฐ์ดํ„ฐ๋ฅผ ์ปฌ๋ ‰์…˜ ๋‹จ์œ„๋กœ ๊ด€๋ฆฌํ•ฉ๋‹ˆ๋‹ค. `collection_name` ํŒŒ๋ผ๋ฏธํ„ฐ๋กœ ์—ฌ๋Ÿฌ ๋ฌธ์„œ ์„ธํŠธ๋ฅผ ๋ถ„๋ฆฌํ•ด ์ €์žฅํ•˜๊ณ  ๊ฐœ๋ณ„๋กœ ๊ฒ€์ƒ‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
      "**๋ฉ”ํƒ€๋ฐ์ดํ„ฐ ํ•„ํ„ฐ๋ง**: `db.similarity_search(query, filter={\"source\": \"file.pdf\"})`์ฒ˜๋Ÿผ ๋ฉ”ํƒ€๋ฐ์ดํ„ฐ ์กฐ๊ฑด์œผ๋กœ ๊ฒ€์ƒ‰ ๋ฒ”์œ„๋ฅผ ์ œํ•œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. FAISS๋Š” ์ง€์›ํ•˜์ง€ ์•Š๋Š” ๊ธฐ๋Šฅ์ž…๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-sentence-transformers",
    "certification": "Docs Study",
    "title": "sentence-transformers โ€” ๋กœ์ปฌ ์ž„๋ฒ ๋”ฉ",
    "level": "์ค‘๊ธ‰",
    "related_practices": ["docs-embed-1"],
    "summary": "sentence-transformers๋Š” HuggingFace ์ž„๋ฒ ๋”ฉ ๋ชจ๋ธ์„ ๋กœ์ปฌ์—์„œ ์‹คํ–‰ํ•ฉ๋‹ˆ๋‹ค. OpenAI API ์—†์ด ์™„์ „ ๋กœ์ปฌ RAG๋ฅผ ๊ตฌ์„ฑํ•  ์ˆ˜ ์žˆ๊ณ , ํ•œ๊ตญ์–ด ํŠนํ™” ๋ชจ๋ธ๋„ ์„ ํƒํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
    "example": "from langchain_huggingface import HuggingFaceEmbeddings\n\nembeddings = HuggingFaceEmbeddings(\n    model_name='sentence-transformers/all-MiniLM-L6-v2'\n)\n\n# ์ž„๋ฒ ๋”ฉ ๋ฒกํ„ฐ ์ƒ์„ฑ\nvector = embeddings.embed_query('ํŒŒ์ด์ฌ์ด ๋ญ์•ผ?')\nprint(len(vector))  # 384์ฐจ์›\n\n# Chroma์™€ ํ•จ๊ป˜\ndb = Chroma.from_documents(chunks, embeddings, persist_directory='./db')",
    "common_mistake": "์ฒซ ์‹คํ–‰ ์‹œ ๋ชจ๋ธ ํŒŒ์ผ์„ ๋‹ค์šด๋กœ๋“œํ•˜๋ฏ€๋กœ ๋А๋ฆฝ๋‹ˆ๋‹ค. Streamlit ์•ฑ์—์„œ๋Š” `@st.cache_resource`๋กœ ๊ฐ์‹ธ ํ•œ ๋ฒˆ๋งŒ ๋กœ๋“œํ•˜์„ธ์š”. ์ž„๋ฒ ๋”ฉ ๋ชจ๋ธ์„ ์ค‘๊ฐ„์— ๋ฐ”๊พธ๋ฉด ๊ธฐ์กด VectorStore์™€ ํ˜ธํ™˜๋˜์ง€ ์•Š์•„ ์ „์ฒด ์žฌ์ธ๋ฑ์‹ฑ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["HuggingFaceEmbeddings", "sentence-transformers", "all-MiniLM-L6-v2", "embed_query", "๋กœ์ปฌ ์ž„๋ฒ ๋”ฉ", "384์ฐจ์›"],
    "source_id": "official-docs-huggingface",
    "details": [
      "**๋ชจ๋ธ ์„ ํƒ**: `all-MiniLM-L6-v2`๋Š” ๋น ๋ฅด๊ณ  ๊ฒฝ๋Ÿ‰(384์ฐจ์›), `all-mpnet-base-v2`๋Š” ํ’ˆ์งˆ์ด ๋” ๋†’์ง€๋งŒ ๋А๋ฆฝ๋‹ˆ๋‹ค. ํ•œ๊ตญ์–ด๊ฐ€ ๋งŽ๋‹ค๋ฉด `jhgan/ko-sroberta-multitask` ๊ฐ™์€ ํ•œ๊ตญ์–ด ํŠนํ™” ๋ชจ๋ธ์„ ์”๋‹ˆ๋‹ค.",
      "**OpenAI vs ๋กœ์ปฌ**: OpenAI `text-embedding-3-small`์€ ํ’ˆ์งˆ์ด ๋†’์ง€๋งŒ API ๋น„์šฉ์ด ๋ฐœ์ƒํ•ฉ๋‹ˆ๋‹ค. ๋กœ์ปฌ ์ž„๋ฒ ๋”ฉ์€ ๋ฌด๋ฃŒ์ด๋‚˜ GPU ์—†์ด๋Š” ๋Œ€์šฉ๋Ÿ‰์—์„œ ๋А๋ฆฝ๋‹ˆ๋‹ค.",
      "**์ฐจ์› ๊ณ ์ • ์ฃผ์˜**: ํ•œ ๋ฒˆ ์ž„๋ฒ ๋”ฉํ•œ ๋ชจ๋ธ์„ ๋ฐ”๊พธ๋ฉด ๊ธฐ์กด ์ธ๋ฑ์Šค์™€ ํ˜ธํ™˜๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ์ƒˆ ๋ชจ๋ธ๋กœ ์ „์ฒด ์ธ๋ฑ์Šค๋ฅผ ์žฌ์ƒ์„ฑํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-ragas",
    "certification": "Docs Study",
    "title": "Ragas โ€” RAG ํ’ˆ์งˆ ์ž๋™ ํ‰๊ฐ€",
    "level": "์ค‘๊ธ‰",
    "related_practices": ["docs-ragas-1", "docs-ragas-2"],
    "summary": "Ragas๋Š” RAG ์‹œ์Šคํ…œ์˜ ํ’ˆ์งˆ์„ ์ž๋™์œผ๋กœ ์ˆ˜์น˜ํ™”ํ•ฉ๋‹ˆ๋‹ค. Faithfulness(ํ™˜๊ฐ ์—†๋Š”๊ฐ€), Answer Relevancy(์งˆ๋ฌธ๊ณผ ๋งž๋Š” ๋‹ต์ธ๊ฐ€), Context Precision(๊ด€๋ จ ๋ฌธ์„œ๋ฅผ ์ž˜ ์ฐพ์•˜๋Š”๊ฐ€)์„ LLM-as-judge๋กœ ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from ragas import evaluate\nfrom ragas.metrics import faithfulness, answer_relevancy, context_precision\nfrom datasets import Dataset\n\ndata = {\n    'question': ['LangChain์ด ๋ญ์•ผ?'],\n    'answer': ['LangChain์€ LLM ์•ฑ ํ”„๋ ˆ์ž„์›Œํฌ์ž…๋‹ˆ๋‹ค.'],\n    'contexts': [['LangChain์€ LLM ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜ ๊ฐœ๋ฐœ์„ ์œ„ํ•œ ํ”„๋ ˆ์ž„์›Œํฌ์ž…๋‹ˆ๋‹ค.']],\n    'ground_truth': ['LangChain์€ LLM ์•ฑ ๊ฐœ๋ฐœ ํ”„๋ ˆ์ž„์›Œํฌ์ž…๋‹ˆ๋‹ค.']\n}\n\ndataset = Dataset.from_dict(data)\nresult = evaluate(dataset, metrics=[faithfulness, answer_relevancy, context_precision])\nprint(result)",
    "common_mistake": "Ragas๋Š” ํ‰๊ฐ€์— LLM์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ๊ธฐ๋ณธ์œผ๋กœ OpenAI GPT๋ฅผ ์”๋‹ˆ๋‹ค. `evaluate(..., llm=your_llm, embeddings=your_embeddings)`๋กœ ๋‹ค๋ฅธ ๋ชจ๋ธ์„ ์ง€์ •ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
    "keywords": ["ragas", "evaluate", "faithfulness", "answer_relevancy", "context_precision", "LLM-as-judge", "Dataset"],
    "source_id": "official-docs-ragas",
    "details": [
      "**3๊ฐ€์ง€ ํ•ต์‹ฌ ๋ฉ”ํŠธ๋ฆญ**: Faithfulness(๋‹ต๋ณ€์ด ๊ฒ€์ƒ‰๋œ ์ปจํ…์ŠคํŠธ์— ๊ทผ๊ฑฐํ•˜๋Š”๊ฐ€ โ€” ํ™˜๊ฐ ํƒ์ง€), Answer Relevancy(๋‹ต๋ณ€์ด ์งˆ๋ฌธ๊ณผ ๊ด€๋ จ ์žˆ๋Š”๊ฐ€), Context Precision(๊ด€๋ จ ๋ฌธ์„œ๊ฐ€ ์ƒ์œ„ k๊ฐœ ์•ˆ์— ์žˆ๋Š”๊ฐ€).",
      "**MLflow ์—ฐ๋™**: `result.to_pandas()`๋กœ DataFrame ๋ณ€ํ™˜ ํ›„ `mlflow.log_metrics()`๋กœ ๊ธฐ๋กํ•˜๋ฉด chunk_size, k ํŒŒ๋ผ๋ฏธํ„ฐ๋ณ„ ํ’ˆ์งˆ ๋ณ€ํ™”๋ฅผ ์ถ”์ ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
      "**ground_truth ์—†์ด๋„ ๊ฐ€๋Šฅ**: `ground_truth`๊ฐ€ ์—†์œผ๋ฉด faithfulness, answer_relevancy๋งŒ ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค. Answer Correctness ๋ฉ”ํŠธ๋ฆญ์€ ground_truth๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-summarization",
    "certification": "Docs Study",
    "title": "LangChain ์š”์•ฝ ์ฒด์ธ โ€” map_reduce & stuff",
    "level": "์ค‘๊ธ‰",
    "related_practices": ["docs-news-1", "docs-news-2"],
    "summary": "LangChain์˜ `load_summarize_chain`์€ ๊ธด ๋ฌธ์„œ๋ฅผ LLM ์ปจํ…์ŠคํŠธ ํ•œ๊ณ„ ๋‚ด์—์„œ ์š”์•ฝํ•˜๋Š” ์ฒด์ธ์„ ๋งŒ๋“ญ๋‹ˆ๋‹ค. ๋ฌธ์„œ๊ฐ€ ์งง์œผ๋ฉด `stuff`, ๊ธธ๋ฉด `map_reduce` ๋ฐฉ์‹์„ ์”๋‹ˆ๋‹ค.",
    "example": "from langchain.chains.summarize import load_summarize_chain\nfrom langchain_core.documents import Document\n\n# map_reduce: ๊ฐ ์ฒญํฌ ์š”์•ฝ โ†’ ์ตœ์ข… ์š”์•ฝ\nchain = load_summarize_chain(llm, chain_type='map_reduce')\n\ndocs = [Document(page_content=article) for article in articles]\nresult = chain.invoke(docs)\nprint(result['output_text'])",
    "common_mistake": "`stuff` ๋ฐฉ์‹์€ ๋ชจ๋“  ๋ฌธ์„œ๋ฅผ ํ•˜๋‚˜์˜ ํ”„๋กฌํ”„ํŠธ์— ๋„ฃ์Šต๋‹ˆ๋‹ค. ๋ฌธ์„œ๊ฐ€ LLM ์ปจํ…์ŠคํŠธ ์ฐฝ๋ณด๋‹ค ํฌ๋ฉด ์˜ค๋ฅ˜๊ฐ€ ๋‚ฉ๋‹ˆ๋‹ค. ์—ฌ๋Ÿฌ ๋ฌธ์„œ๋‚˜ ๊ธด ๋ฌธ์„œ๋Š” ํ•ญ์ƒ `map_reduce`๋ฅผ ๋จผ์ € ๊ณ ๋ คํ•˜์„ธ์š”.",
    "keywords": ["load_summarize_chain", "map_reduce", "stuff", "refine", "output_text", "์š”์•ฝ"],
    "source_id": "official-docs-langchain",
    "details": [
      "**chain_type ๋น„๊ต**: `stuff`(์ „์ฒด ํ•œ ๋ฒˆ), `map_reduce`(์ฒญํฌ๋ณ„ ์š”์•ฝ ํ›„ ํ•ฉ์นจ), `refine`(์ด์ „ ์š”์•ฝ์„ ๋‹ค์Œ ์ฒญํฌ์— ๋ฐ˜์˜). ํ’ˆ์งˆ์€ refine > map_reduce > stuff์ด์ง€๋งŒ ์†๋„๋Š” ๋ฐ˜๋Œ€์ž…๋‹ˆ๋‹ค.",
      "**map_prompt / combine_prompt**: `load_summarize_chain`์— `map_prompt`์™€ `combine_prompt`๋ฅผ ์ง€์ •ํ•ด ์š”์•ฝ ์ง€์‹œ๋ฌธ์„ ์ปค์Šคํ„ฐ๋งˆ์ด์ง•ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
      "**๋น„๋™๊ธฐ**: `chain.ainvoke(docs)`๋กœ ๋น„๋™๊ธฐ ์š”์ฒญ์ด ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค. ์—ฌ๋Ÿฌ ๋‰ด์Šค ๊ธฐ์‚ฌ๋ฅผ ๋ณ‘๋ ฌ ์š”์•ฝํ•  ๋•Œ `asyncio.gather`์™€ ํ•จ๊ป˜ ์“ฐ๋ฉด ์†๋„๊ฐ€ ํฌ๊ฒŒ ๊ฐœ์„ ๋ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-sql-chain",
    "certification": "Docs Study",
    "title": "LangChain Text-to-SQL โ€” ์ž์—ฐ์–ด๋กœ DB ์กฐํšŒ",
    "level": "์ค‘๊ธ‰",
    "related_practices": ["docs-sql-1", "docs-sql-2"],
    "summary": "LangChain์˜ `create_sql_query_chain`์€ ์ž์—ฐ์–ด ์งˆ๋ฌธ์„ SQL ์ฟผ๋ฆฌ๋กœ ๋ณ€ํ™˜ํ•ฉ๋‹ˆ๋‹ค. `SQLDatabase`๋กœ DB ์Šคํ‚ค๋งˆ๋ฅผ ์ž๋™์œผ๋กœ ์ฝ์–ด LLM์—๊ฒŒ ์ปจํ…์ŠคํŠธ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from langchain_community.utilities import SQLDatabase\nfrom langchain.chains import create_sql_query_chain\n\ndb = SQLDatabase.from_uri('sqlite:///./mydb.db')\nchain = create_sql_query_chain(llm, db)\n\nsql = chain.invoke({'question': '๊ฐ€์žฅ ๋งŽ์ด ํŒ”๋ฆฐ ์ƒํ’ˆ 5๊ฐœ๋Š”?'})\nprint(sql)  # SELECT product, SUM(quantity) ...\n\nresult = db.run(sql)\nprint(result)",
    "common_mistake": "LLM์ด ์ƒ์„ฑํ•œ SQL์„ ๋ฐ”๋กœ `db.run()`์— ๋„˜๊ธฐ๋ฉด ์ž˜๋ชป๋œ ์ฟผ๋ฆฌ๊ฐ€ ์‹คํ–‰๋ฉ๋‹ˆ๋‹ค. Guardrails์˜ `ValidSQL`์ด๋‚˜ ๋ณ„๋„ ํŒŒ์‹ฑ์œผ๋กœ ๊ฒ€์ฆํ•œ ๋’ค ์‹คํ–‰ํ•˜์„ธ์š”.",
    "keywords": ["SQLDatabase", "create_sql_query_chain", "from_uri", "db.run", "Text-to-SQL", "sqlite"],
    "source_id": "official-docs-langchain",
    "details": [
      "**์Šคํ‚ค๋งˆ ์ž๋™ ์ฃผ์ž…**: `SQLDatabase`๋Š” ์—ฐ๊ฒฐ๋œ DB์˜ ํ…Œ์ด๋ธ”๋ช…๊ณผ ์ปฌ๋Ÿผ ์ •๋ณด๋ฅผ ์ž๋™์œผ๋กœ ํ”„๋กฌํ”„ํŠธ์— ์ฃผ์ž…ํ•ฉ๋‹ˆ๋‹ค. LLM์ด ์กด์žฌํ•˜๋Š” ํ…Œ์ด๋ธ”/์ปฌ๋Ÿผ๋งŒ ์ฐธ์กฐํ•˜๋„๋ก ์œ ๋„ํ•ฉ๋‹ˆ๋‹ค.",
      "**QuerySQLDatabaseTool**: SQL ์ƒ์„ฑ๊ณผ ์‹คํ–‰์„ ํ•˜๋‚˜์˜ ํˆด๋กœ ๋ฌถ์€ ๊ณ ์ˆ˜์ค€ ์ธํ„ฐํŽ˜์ด์Šค. Agent์— ๋“ฑ๋กํ•˜๋ฉด LLM์ด ํ•„์š”ํ•  ๋•Œ DB๋ฅผ ์ง์ ‘ ์กฐํšŒํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
      "**๋ณด์•ˆ ์ฃผ์˜**: `db.run()`์€ ์‹ค์ œ DB๋ฅผ ์ˆ˜์ •ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ฝ๊ธฐ ์ „์šฉ ์‚ฌ์šฉ์ž ๊ณ„์ •์œผ๋กœ ์—ฐ๊ฒฐํ•˜๊ฑฐ๋‚˜ SELECT๋งŒ ํ—ˆ์šฉํ•˜๋Š” wrapper๋ฅผ ๋‘๋Š” ๊ฒƒ์ด ์•ˆ์ „ํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-langgraph",
    "certification": "Docs Study",
    "title": "LangGraph โ€” ์ƒํƒœ ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ ์›Œํฌํ”Œ๋กœ",
    "level": "๊ณ ๊ธ‰",
    "related_practices": ["docs-graph-1", "docs-graph-2"],
    "summary": "LangGraph๋Š” LLM ์—์ด์ „ํŠธ๋ฅผ ๊ทธ๋ž˜ํ”„(๋…ธ๋“œ + ์—ฃ์ง€)๋กœ ์ •์˜ํ•ฉ๋‹ˆ๋‹ค. ๋‹จ์ˆœ ์ฒด์ธ์˜ Aโ†’Bโ†’C์™€ ๋‹ฌ๋ฆฌ ์กฐ๊ฑด ๋ถ„๊ธฐ, ๋ฃจํ”„, ๋ณ‘๋ ฌ ์‹คํ–‰์ด ๊ฐ€๋Šฅํ•ด ์‹ค์ œ ์—์ด์ „ํŠธ ํ–‰๋™์„ ํ‘œํ˜„ํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from langgraph.graph import StateGraph, END\nfrom typing import TypedDict\n\nclass State(TypedDict):\n    question: str\n    answer: str\n\ndef retrieve(state: State) -> State:\n    return {\"answer\": str(retriever.invoke(state[\"question\"]))}\n\ndef generate(state: State) -> State:\n    return {\"answer\": chain.invoke(state)}\n\ngraph = StateGraph(State)\ngraph.add_node(\"retrieve\", retrieve)\ngraph.add_node(\"generate\", generate)\ngraph.set_entry_point(\"retrieve\")\ngraph.add_edge(\"retrieve\", \"generate\")\ngraph.add_edge(\"generate\", END)\n\napp = graph.compile()\nresult = app.invoke({\"question\": \"LangChain์ด ๋ญ์•ผ?\"})",
    "common_mistake": "LangGraph๋Š” LangChain ์ฒด์ธ์ด ์•„๋‹™๋‹ˆ๋‹ค. `graph.compile()` ํ›„ `app.invoke()`๋ฅผ ์”๋‹ˆ๋‹ค. State TypedDict๋ฅผ ์ •ํ™•ํžˆ ์ •์˜ํ•˜์ง€ ์•Š์œผ๋ฉด ๋…ธ๋“œ์—์„œ ํ‚ค ์˜ค๋ฅ˜๊ฐ€ ๋ฐœ์ƒํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["LangGraph", "StateGraph", "State", "node", "edge", "compile", "conditional_edges", "END"],
    "source_id": "official-docs-langgraph",
    "details": [
      "**๋…ธ๋“œ vs ์ฒด์ธ**: ์ฒด์ธ์€ Aโ†’Bโ†’C ์„ ํ˜•. ๊ทธ๋ž˜ํ”„๋Š” A์—์„œ ์กฐ๊ฑด์— ๋”ฐ๋ผ B ๋˜๋Š” C๋กœ ๋ถ„๊ธฐ, D์—์„œ A๋กœ ๋ฃจํ”„ ๋“ฑ ๋ณต์žกํ•œ ํ๋ฆ„์„ ํ‘œํ˜„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
      "**conditional_edges**: `graph.add_conditional_edges('judge', router_fn, {'retry': 'retrieve', 'done': END})`์ฒ˜๋Ÿผ ๋…ธ๋“œ ๋ฐ˜ํ™˜๊ฐ’์— ๋”ฐ๋ผ ๋‹ค์Œ ๋…ธ๋“œ๋ฅผ ๋™์ ์œผ๋กœ ๊ฒฐ์ •ํ•ฉ๋‹ˆ๋‹ค. ReAct ์—์ด์ „ํŠธ์˜ ํ•ต์‹ฌ์ž…๋‹ˆ๋‹ค.",
      "**checkpointer**: `MemorySaver`๋ฅผ `graph.compile(checkpointer=...)`์— ๋„˜๊ธฐ๋ฉด ์‹คํ–‰ ์ƒํƒœ๋ฅผ ์ €์žฅํ•ด ์ค‘๋‹จ ํ›„ ์žฌ๊ฐœ, ๋ฉ€ํ‹ฐํ„ด ๋Œ€ํ™” ์œ ์ง€๊ฐ€ ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-airflow",
    "certification": "Docs Study",
    "title": "Apache Airflow โ€” ์›Œํฌํ”Œ๋กœ ์˜ค์ผ€์ŠคํŠธ๋ ˆ์ด์…˜",
    "level": "๊ณ ๊ธ‰",
    "related_practices": ["docs-lab-23"],
    "summary": "Airflow๋Š” DAG(Directed Acyclic Graph)๋กœ ๋ฐ์ดํ„ฐ ํŒŒ์ดํ”„๋ผ์ธ์„ ์ •์˜ํ•˜๊ณ  ์Šค์ผ€์ค„๋งยท๋ชจ๋‹ˆํ„ฐ๋งํ•ฉ๋‹ˆ๋‹ค. ML ํŒŒ์ดํ”„๋ผ์ธ(๋ฐ์ดํ„ฐ ์ˆ˜์ง‘ โ†’ ์ „์ฒ˜๋ฆฌ โ†’ ํ•™์Šต โ†’ ๋ฐฐํฌ)์ด๋‚˜ ๋ฌธ์„œ ์žฌ์ธ๋ฑ์‹ฑ ์ž๋™ํ™”์— ์”๋‹ˆ๋‹ค.",
    "example": "from airflow import DAG\nfrom airflow.operators.python import PythonOperator\nfrom datetime import datetime\n\ndef reindex():\n    # ๋ฌธ์„œ ๋กœ๋“œ โ†’ ์ž„๋ฒ ๋”ฉ โ†’ VectorStore ์—…๋ฐ์ดํŠธ\n    pass\n\nwith DAG('daily_reindex', schedule='@daily', start_date=datetime(2024,1,1)) as dag:\n    task = PythonOperator(task_id='reindex', python_callable=reindex)",
    "common_mistake": "DAG ํŒŒ์ผ์€ Airflow ์Šค์ผ€์ค„๋Ÿฌ๊ฐ€ ์ฃผ๊ธฐ์ ์œผ๋กœ ํŒŒ์‹ฑํ•ฉ๋‹ˆ๋‹ค. ํŒŒ์ผ ์ตœ์ƒ์œ„์— ๋ฌด๊ฑฐ์šด ์—ฐ์‚ฐ(DB ์—ฐ๊ฒฐ, ๋ชจ๋ธ ๋กœ๋“œ)์„ ๋‘๋ฉด ํŒŒ์‹ฑ๋งˆ๋‹ค ์‹คํ–‰๋ผ ์Šค์ผ€์ค„๋Ÿฌ๊ฐ€ ๋А๋ ค์ง‘๋‹ˆ๋‹ค. ๋ฐ˜๋“œ์‹œ task ํ•จ์ˆ˜ ์•ˆ์— ๋„ฃ์œผ์„ธ์š”.",
    "keywords": ["airflow", "DAG", "task", "operator", "schedule", "XCom", "pipeline", "orchestration"],
    "source_id": "official-docs-airflow",
    "details": [
      "**Operator ์ข…๋ฅ˜**: `PythonOperator`(ํŒŒ์ด์ฌ ํ•จ์ˆ˜), `BashOperator`(์…ธ ๋ช…๋ น), `DockerOperator`(์ปจํ…Œ์ด๋„ˆ ์‹คํ–‰) ๋“ฑ. ML ํŒŒ์ดํ”„๋ผ์ธ์—” ์ฃผ๋กœ PythonOperator๋ฅผ ์”๋‹ˆ๋‹ค.",
      "**XCom**: task ๊ฐ„ ๋ฐ์ดํ„ฐ ์ „๋‹ฌ ๋ฉ”์ปค๋‹ˆ์ฆ˜. `ti.xcom_push(key, value)` / `ti.xcom_pull(task_ids, key)`๋กœ ์†Œ๋Ÿ‰ ๋ฐ์ดํ„ฐ๋ฅผ ๊ณต์œ ํ•ฉ๋‹ˆ๋‹ค. ๋Œ€์šฉ๋Ÿ‰์€ S3/GCS ๊ฒฝ๋กœ๋ฅผ ์ „๋‹ฌํ•˜๋Š” ๋ฐฉ์‹์„ ์”๋‹ˆ๋‹ค.",
      "**์˜์กด์„ฑ**: `task_a >> task_b >> task_c` ๋˜๋Š” `task_a.set_downstream(task_b)` ๋กœ ์‹คํ–‰ ์ˆœ์„œ๋ฅผ ์ •์˜ํ•ฉ๋‹ˆ๋‹ค. ๋ณ‘๋ ฌ ์‹คํ–‰์€ `[task_b, task_c] >> task_d` ํ˜•ํƒœ๋กœ ํ‘œํ˜„ํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-mlflow",
    "certification": "Docs Study",
    "title": "MLflow โ€” ์‹คํ—˜ ์ถ”์  & ๋ชจ๋ธ ๊ด€๋ฆฌ",
    "level": "๊ณ ๊ธ‰",
    "related_practices": ["docs-lab-15"],
    "summary": "MLflow๋Š” ML ์‹คํ—˜์˜ ํŒŒ๋ผ๋ฏธํ„ฐ, ๋ฉ”ํŠธ๋ฆญ, ์•„ํ‹ฐํŒฉํŠธ๋ฅผ ์ถ”์ ํ•˜๊ณ  ๋ชจ๋ธ์„ ๋ฒ„์ „ ๊ด€๋ฆฌํ•ฉ๋‹ˆ๋‹ค. RAG์—์„œ chunk_size, k, ์ž„๋ฒ ๋”ฉ ๋ชจ๋ธ ๋“ฑ ํŒŒ๋ผ๋ฏธํ„ฐ ์‹คํ—˜ ๊ฒฐ๊ณผ๋ฅผ ์ฒด๊ณ„์ ์œผ๋กœ ๋น„๊ตํ•  ๋•Œ ์”๋‹ˆ๋‹ค.",
    "example": "import mlflow\n\nwith mlflow.start_run(run_name='rag-exp-1'):\n    mlflow.log_param('chunk_size', 500)\n    mlflow.log_param('k', 4)\n    mlflow.log_metric('answer_relevance', 0.87)\n    mlflow.log_metric('faithfulness', 0.92)\n    mlflow.log_artifact('eval_results.json')",
    "common_mistake": "run์„ `with mlflow.start_run()` ๋ธ”๋ก์œผ๋กœ ๊ฐ์‹ธ์ง€ ์•Š์œผ๋ฉด ์‹ค์ˆ˜๋กœ ์ด์ „ run์— ๋กœ๊น…๋˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ํ•ญ์ƒ context manager๋ฅผ ์‚ฌ์šฉํ•˜์„ธ์š”.",
    "keywords": ["mlflow", "experiment", "run", "log_param", "log_metric", "artifact", "model registry"],
    "source_id": "official-docs-mlflow",
    "details": [
      "**UI**: `mlflow ui` ๋ช…๋ น์œผ๋กœ localhost:5000์—์„œ ์‹คํ—˜ ๋น„๊ต ๋Œ€์‹œ๋ณด๋“œ๋ฅผ ์‹คํ–‰ํ•ฉ๋‹ˆ๋‹ค. ํŒŒ๋ผ๋ฏธํ„ฐ๋ณ„ ๋ฉ”ํŠธ๋ฆญ ์ถ”์ด๋ฅผ ์‹œ๊ฐ์ ์œผ๋กœ ๋น„๊ตํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
      "**Model Registry**: `mlflow.log_model()`๋กœ ๋ชจ๋ธ์„ ์ €์žฅํ•˜๊ณ  `Staging`โ†’`Production` ๋‹จ๊ณ„๋ฅผ ๊ด€๋ฆฌํ•ฉ๋‹ˆ๋‹ค. ๋ฐฐํฌ ์‹œ `mlflow.pyfunc.load_model('models:/name/Production')`์œผ๋กœ ๋กœ๋“œํ•ฉ๋‹ˆ๋‹ค.",
      "**LangChain ํ†ตํ•ฉ**: `mlflow.langchain.autolog()`๋ฅผ ํ™œ์„ฑํ™”ํ•˜๋ฉด ์ฒด์ธ ์‹คํ–‰๋งˆ๋‹ค ํŒŒ๋ผ๋ฏธํ„ฐ์™€ ๋ฉ”ํŠธ๋ฆญ์ด ์ž๋™ ๊ธฐ๋ก๋ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-mcp",
    "certification": "Docs Study",
    "title": "MCP โ€” Model Context Protocol",
    "level": "๊ณ ๊ธ‰",
    "related_practices": ["docs-lab-24"],
    "summary": "MCP(Model Context Protocol)๋Š” LLM ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์ด ์™ธ๋ถ€ ๋„๊ตฌยท๋ฐ์ดํ„ฐ์†Œ์Šค์™€ ํ‘œ์ค€ํ™”๋œ ๋ฐฉ์‹์œผ๋กœ ์—ฐ๊ฒฐํ•˜๋Š” Anthropic ์˜คํ”ˆ ํ”„๋กœํ† ์ฝœ์ž…๋‹ˆ๋‹ค. Claude Desktop์ด๋‚˜ Claude Code์— ๋กœ์ปฌ MCP ์„œ๋ฒ„๋ฅผ ์—ฐ๊ฒฐํ•˜๋ฉด ํŒŒ์ผ ์‹œ์Šคํ…œ, DB, ์™ธ๋ถ€ API๋ฅผ LLM์ด ์ง์ ‘ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.",
    "example": "# mcp_server.py (FastMCP ์‚ฌ์šฉ)\nfrom mcp.server.fastmcp import FastMCP\n\nmcp = FastMCP('my-server')\n\n@mcp.tool()\ndef search_docs(query: str) -> str:\n    '''๋ฒกํ„ฐ DB์—์„œ ๊ด€๋ จ ๋ฌธ์„œ๋ฅผ ๊ฒ€์ƒ‰ํ•ฉ๋‹ˆ๋‹ค.'''\n    results = retriever.invoke(query)\n    return '\\n'.join(r.page_content for r in results)\n\nmcp.run(transport='stdio')",
    "common_mistake": "MCP ์„œ๋ฒ„๋Š” stdio ๋˜๋Š” SSE transport ์ค‘ ํ•˜๋‚˜๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค. Claude Desktop์€ stdio(๋กœ์ปฌ ํ”„๋กœ์„ธ์Šค), Claude Code๋Š” SSE(HTTP ์„œ๋ฒ„) ๋ฐฉ์‹์„ ์ฃผ๋กœ ์”๋‹ˆ๋‹ค. transport ๋ถˆ์ผ์น˜ ์‹œ ์—ฐ๊ฒฐ์ด ์•ˆ ๋ฉ๋‹ˆ๋‹ค.",
    "keywords": ["MCP", "Model Context Protocol", "tool", "resource", "FastMCP", "stdio", "Claude", "agent"],
    "source_id": "official-docs-mcp",
    "details": [
      "**3๊ฐ€์ง€ ๊ธฐ๋ณธ ์š”์†Œ**: `tool`(LLM์ด ํ˜ธ์ถœํ•˜๋Š” ํ•จ์ˆ˜), `resource`(LLM์ด ์ฝ๋Š” ๋ฐ์ดํ„ฐ), `prompt`(์žฌ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•œ ํ”„๋กฌํ”„ํŠธ ํ…œํ”Œ๋ฆฟ). ๋Œ€๋ถ€๋ถ„์˜ ์„œ๋ฒ„๋Š” `tool`๋งŒ ๊ตฌํ˜„ํ•ฉ๋‹ˆ๋‹ค.",
      "**Claude Desktop ์—ฐ๊ฒฐ**: `~/Library/Application Support/Claude/claude_desktop_config.json`์— ์„œ๋ฒ„ ๊ฒฝ๋กœ์™€ ๋ช…๋ น์–ด๋ฅผ ๋“ฑ๋กํ•˜๋ฉด ๋Œ€ํ™” ์ค‘ ์ž๋™์œผ๋กœ ๋„๊ตฌ๊ฐ€ ๋ณด์ž…๋‹ˆ๋‹ค.",
      "**FastMCP**: Anthropic ๊ณต์‹ Python SDK์˜ ๊ณ ์ˆ˜์ค€ ๋ž˜ํผ. `@mcp.tool()` ๋ฐ์ฝ”๋ ˆ์ดํ„ฐ๋กœ ํ•จ์ˆ˜๋ฅผ ๋„๊ตฌ๋กœ ๋“ฑ๋กํ•˜๋ฉด ์Šคํ‚ค๋งˆ ์ƒ์„ฑ, ์ž…์ถœ๋ ฅ ๊ฒ€์ฆ์ด ์ž๋™์œผ๋กœ ์ฒ˜๋ฆฌ๋ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-litellm",
    "certification": "Docs Study",
    "title": "LiteLLM โ€” ์—ฌ๋Ÿฌ LLM์„ ํ•˜๋‚˜์˜ ์ธํ„ฐํŽ˜์ด์Šค๋กœ",
    "level": "๊ณ ๊ธ‰",
    "related_practices": ["docs-litellm-1"],
    "summary": "LiteLLM์€ OpenAI, Anthropic, Ollama, Gemini ๋“ฑ 100๊ฐœ ์ด์ƒ์˜ LLM ํ”„๋กœ๋ฐ”์ด๋”๋ฅผ ๋™์ผํ•œ API ํ˜•์‹์œผ๋กœ ํ˜ธ์ถœํ•ฉ๋‹ˆ๋‹ค. ๋ชจ๋ธ์„ ๋ฐ”๊ฟ€ ๋•Œ ์ฝ”๋“œ ์ˆ˜์ • ์—†์ด ๋ชจ๋ธ๋ช…๋งŒ ๋ณ€๊ฒฝํ•˜๋ฉด ๋ฉ๋‹ˆ๋‹ค.",
    "example": "from litellm import completion\n\n# Ollama ๋กœ์ปฌ\nresponse = completion(\n    model='ollama/llama3.2',\n    messages=[{'role': 'user', 'content': '์•ˆ๋…•'}]\n)\n\n# OpenAI โ€” ์ฝ”๋“œ ๋™์ผ, ๋ชจ๋ธ๋ช…๋งŒ ๋ณ€๊ฒฝ\nresponse = completion(\n    model='gpt-4o-mini',\n    messages=[{'role': 'user', 'content': '์•ˆ๋…•'}]\n)\n\nprint(response.choices[0].message.content)",
    "common_mistake": "๋ชจ๋ธ๋ช… ํ˜•์‹์€ `'ํ”„๋กœ๋ฐ”์ด๋”/๋ชจ๋ธ๋ช…'`์ž…๋‹ˆ๋‹ค. Ollama๋Š” `'ollama/llama3.2'`, Anthropic์€ `'anthropic/claude-3-5-sonnet-20241022'`. OpenAI๋งŒ ์˜ˆ์™ธ๋กœ ์ ‘๋‘์‚ฌ ์—†์ด `'gpt-4o-mini'`์ฒ˜๋Ÿผ ์”๋‹ˆ๋‹ค.",
    "keywords": ["LiteLLM", "completion", "model", "provider", "๋ฒค๋” ๋…๋ฆฝ", "OpenAI ํ˜ธํ™˜", "ChatLiteLLM"],
    "source_id": "official-docs-litellm",
    "details": [
      "**๋น„์šฉ ์ถ”์ **: `litellm.success_callback`์œผ๋กœ ๊ฐ ํ˜ธ์ถœ์˜ ํ† ํฐ ์ˆ˜์™€ ์˜ˆ์ƒ ๋น„์šฉ์„ ์ž๋™ ๊ธฐ๋กํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๋Ÿฌ ๋ชจ๋ธ์„ ์‹คํ—˜ํ•  ๋•Œ ๋น„์šฉ ๋น„๊ต์— ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค.",
      "**LangChain ํ†ตํ•ฉ**: `from langchain_community.chat_models import ChatLiteLLM`์œผ๋กœ LangChain ์ฒด์ธ์—์„œ LiteLLM์„ ์”๋‹ˆ๋‹ค. LCEL ํŒŒ์ดํ”„๋ผ์ธ์— ๋ฐ”๋กœ ์—ฐ๊ฒฐ๋ฉ๋‹ˆ๋‹ค.",
      "**Fallback**: `completion(..., fallbacks=['gpt-4o-mini', 'ollama/llama3.2'])`์ฒ˜๋Ÿผ ์ฃผ ๋ชจ๋ธ ์‹คํŒจ ์‹œ ์ž๋™์œผ๋กœ ๋ฐฑ์—… ๋ชจ๋ธ๋กœ ์ „ํ™˜ํ•˜๋Š” ๊ธฐ๋Šฅ์„ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-chainlit",
    "certification": "Docs Study",
    "title": "Chainlit โ€” ์ฑ—๋ด‡ UI ํŠนํ™” ํ”„๋ ˆ์ž„์›Œํฌ",
    "level": "๊ณ ๊ธ‰",
    "related_practices": ["docs-chainlit-1"],
    "summary": "Chainlit์€ LLM ์ฑ—๋ด‡ UI์— ํŠนํ™”๋œ ํ”„๋ ˆ์ž„์›Œํฌ์ž…๋‹ˆ๋‹ค. Streamlit๊ณผ ๋‹ฌ๋ฆฌ ์›น์†Œ์ผ“ ๊ธฐ๋ฐ˜์œผ๋กœ ๋™์ž‘ํ•ด ์ŠคํŠธ๋ฆฌ๋ฐ ์‘๋‹ต, ํŒŒ์ผ ์—…๋กœ๋“œ, ๋ฉ€ํ‹ฐ์Šคํ… ์ง„ํ–‰ ํ‘œ์‹œ๊ฐ€ ์ž์—ฐ์Šค๋Ÿฝ์Šต๋‹ˆ๋‹ค.",
    "example": "import chainlit as cl\n\n@cl.on_message\nasync def main(message: cl.Message):\n    # ์ŠคํŠธ๋ฆฌ๋ฐ ์‘๋‹ต\n    msg = cl.Message(content='')\n    async for chunk in chain.astream({'question': message.content}):\n        await msg.stream_token(chunk)\n    await msg.send()\n\n# ์‹คํ–‰: chainlit run app.py",
    "common_mistake": "Chainlit์€ `async` ๊ธฐ๋ฐ˜์ž…๋‹ˆ๋‹ค. LangChain์˜ `chain.invoke()`(๋™๊ธฐ)๋ฅผ ์“ฐ๋ฉด ์ด๋ฒคํŠธ ๋ฃจํ”„๊ฐ€ ๋ธ”๋กœํ‚น๋ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋“œ์‹œ `chain.ainvoke()` ๋˜๋Š” `chain.astream()`์„ ์‚ฌ์šฉํ•˜์„ธ์š”.",
    "keywords": ["chainlit", "on_message", "Message", "stream_token", "async", "astream", "on_chat_start"],
    "source_id": "official-docs-chainlit",
    "details": [
      "**๋ฐ์ฝ”๋ ˆ์ดํ„ฐ ๊ธฐ๋ฐ˜**: `@cl.on_chat_start`(์„ธ์…˜ ์‹œ์ž‘ ์‹œ ์ฒด์ธ ์ดˆ๊ธฐํ™”), `@cl.on_message`(๋ฉ”์‹œ์ง€ ์ˆ˜์‹ ), `@cl.on_stop`(์ŠคํŠธ๋ฆฌ๋ฐ ์ค‘๋‹จ)์œผ๋กœ ์ด๋ฒคํŠธ๋ฅผ ์ฒ˜๋ฆฌํ•ฉ๋‹ˆ๋‹ค.",
      "**Step UI**: `async with cl.Step(name='๊ฒ€์ƒ‰ ์ค‘...'):` ๋ธ”๋ก์œผ๋กœ ์—์ด์ „ํŠธ์˜ ๊ฐ ๋‹จ๊ณ„(๊ฒ€์ƒ‰, ์ถ”๋ก  ๋“ฑ)๋ฅผ UI์— ์‹ค์‹œ๊ฐ„์œผ๋กœ ํŽผ์ณ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. LangGraph ์—์ด์ „ํŠธ์™€ ํ•จ๊ป˜ ์“ฐ๋ฉด ๊ฐ•๋ ฅํ•ฉ๋‹ˆ๋‹ค.",
      "**ํŒŒ์ผ ์—…๋กœ๋“œ**: `@cl.on_message`์—์„œ `message.elements`๋กœ ์ฒจ๋ถ€ ํŒŒ์ผ์„ ๋ฐ›์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. PDF ์—…๋กœ๋“œ โ†’ ์ฆ‰์‹œ RAG ์ธ๋ฑ์‹ฑ ํ๋ฆ„์„ ๋ช‡ ์ค„๋กœ ๊ตฌํ˜„ํ•ฉ๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-redis",
    "certification": "Docs Study",
    "title": "Redis โ€” ๋Œ€ํ™” ๋ฉ”๋ชจ๋ฆฌ ์˜์†ํ™” & LLM ์บ์‹ฑ",
    "level": "๊ณ ๊ธ‰",
    "related_practices": ["docs-redis-1"],
    "summary": "Redis๋Š” ์ธ๋ฉ”๋ชจ๋ฆฌ ๋ฐ์ดํ„ฐ ์ €์žฅ์†Œ๋กœ LLM ์•ฑ์—์„œ ๋Œ€ํ™” ํžˆ์Šคํ† ๋ฆฌ ์˜์†ํ™”์™€ LLM ์‘๋‹ต ์บ์‹ฑ์— ์”๋‹ˆ๋‹ค. ์„œ๋ฒ„๋ฅผ ์žฌ์‹œ์ž‘ํ•ด๋„ ๋Œ€ํ™”๊ฐ€ ์œ ์ง€๋˜๊ณ , ๋™์ผ ์งˆ๋ฌธ์— ๋Œ€ํ•œ LLM ํ˜ธ์ถœ ๋น„์šฉ์„ ์ ˆ๊ฐํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from langchain_community.chat_message_histories import RedisChatMessageHistory\nfrom langchain_core.runnables.history import RunnableWithMessageHistory\n\ndef get_session_history(session_id: str):\n    return RedisChatMessageHistory(\n        session_id=session_id,\n        url='redis://localhost:6379'\n    )\n\nchain_with_history = RunnableWithMessageHistory(\n    chain,\n    get_session_history,\n    input_messages_key='question',\n    history_messages_key='chat_history'\n)",
    "common_mistake": "Redis ์„œ๋ฒ„๊ฐ€ ์‹คํ–‰ ์ค‘์ด์ง€ ์•Š์œผ๋ฉด ์—ฐ๊ฒฐ ์˜ค๋ฅ˜๊ฐ€ ๋‚ฉ๋‹ˆ๋‹ค. ๋กœ์ปฌ ๊ฐœ๋ฐœ ์‹œ `docker run -d -p 6379:6379 redis`๋กœ ๋จผ์ € Redis๋ฅผ ์‹คํ–‰ํ•˜์„ธ์š”.",
    "keywords": ["Redis", "RedisChatMessageHistory", "session_id", "url", "์˜์†ํ™”", "์บ์‹ฑ", "TTL"],
    "source_id": "official-docs-redis",
    "details": [
      "**์ธ๋ฉ”๋ชจ๋ฆฌ ๊ต์ฒด**: ๊ธฐ์กด `ChatMessageHistory`(์žฌ์‹œ์ž‘ ์‹œ ์†Œ๋ฉธ)๋ฅผ `RedisChatMessageHistory`๋กœ ๊ต์ฒดํ•˜๋ฉด `get_session_history` ํ•จ์ˆ˜ ํ•œ ์ค„๋งŒ ๋ฐ”๊ฟ”๋„ ์˜์†์„ฑ์„ ์–ป์Šต๋‹ˆ๋‹ค.",
      "**TTL**: `RedisChatMessageHistory(session_id, url, ttl=3600)`์œผ๋กœ 1์‹œ๊ฐ„ ํ›„ ์ž๋™ ๋งŒ๋ฃŒ๋ฉ๋‹ˆ๋‹ค. ๋ฌดํ•œํžˆ ์Œ“์ด๋Š” ํžˆ์Šคํ† ๋ฆฌ์™€ ๋ฉ”๋ชจ๋ฆฌ ๋ˆ„์ˆ˜๋ฅผ ๋ฐฉ์ง€ํ•ฉ๋‹ˆ๋‹ค.",
      "**LLM ์‘๋‹ต ์บ์‹ฑ**: `from langchain.cache import RedisCache; langchain.llm_cache = RedisCache(redis_client)`๋กœ ๋™์ผ ์งˆ๋ฌธ์— ๋Œ€ํ•œ LLM ์‘๋‹ต์„ ์บ์‹œํ•ด ๋น„์šฉ๊ณผ ์‘๋‹ต ์‹œ๊ฐ„์„ ์ค„์ž…๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-unstructured",
    "certification": "Docs Study",
    "title": "Unstructured โ€” PDFยทWordยท์ด๋ฏธ์ง€ ํŒŒ์‹ฑ",
    "level": "๊ณ ๊ธ‰",
    "related_practices": ["docs-unstructured-1"],
    "summary": "Unstructured๋Š” PDF, Word, Excel, PowerPoint, ์ด๋ฏธ์ง€ ๋“ฑ ๋น„์ •ํ˜• ๋ฌธ์„œ๋ฅผ ํ…์ŠคํŠธ๋กœ ๋ณ€ํ™˜ํ•ฉ๋‹ˆ๋‹ค. LangChain์˜ `UnstructuredFileLoader`์™€ ํ†ตํ•ฉ๋˜์–ด ํ™•์žฅ์ž์— ์ƒ๊ด€์—†์ด ๋™์ผ ์ฝ”๋“œ๋กœ RAG ํŒŒ์ดํ”„๋ผ์ธ์— ์—ฐ๊ฒฐํ•ฉ๋‹ˆ๋‹ค.",
    "example": "from langchain_community.document_loaders import UnstructuredFileLoader\n\n# ๋‹จ์ผ ํŒŒ์ผ โ€” ํ™•์žฅ์ž ์ž๋™ ๊ฐ์ง€\nloader = UnstructuredFileLoader('./report.pdf')\ndocs = loader.load()\n\n# ํ˜ผํ•ฉ ํ˜•์‹ ๋””๋ ‰ํ† ๋ฆฌ\nfrom langchain_community.document_loaders import DirectoryLoader\nloader = DirectoryLoader(\n    './docs',\n    glob='**/*',\n    loader_cls=UnstructuredFileLoader\n)\ndocs = loader.load()",
    "common_mistake": "`pip install unstructured`๋Š” ๊ธฐ๋ณธ ์„ค์น˜์ž…๋‹ˆ๋‹ค. PDF์—๋Š” `pip install unstructured[pdf]`, ์ด๋ฏธ์ง€ OCR์—๋Š” `pip install unstructured[local-inference]`์ฒ˜๋Ÿผ ๋ณ„๋„ extra๋ฅผ ์„ค์น˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["Unstructured", "UnstructuredFileLoader", "PDF", "Word", "OCR", "๋น„์ •ํ˜• ๋ฌธ์„œ", "elements"],
    "source_id": "official-docs-unstructured",
    "details": [
      "**์ž๋™ ํŒŒ์ผ ๊ฐ์ง€**: ํ™•์žฅ์ž์— ๋”ฐ๋ผ ์ž๋™์œผ๋กœ ์ ์ ˆํ•œ ํŒŒ์„œ๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค. `.pdf`, `.docx`, `.pptx`, `.jpg`, `.eml` ๋“ฑ์„ ๋™์ผ ์ฝ”๋“œ๋กœ ์ฒ˜๋ฆฌํ•ฉ๋‹ˆ๋‹ค.",
      "**์š”์†Œ ๋ถ„๋ฅ˜**: `mode='elements'`๋กœ ๋กœ๋“œํ•˜๋ฉด Title, NarrativeText, Table, Image ๋“ฑ ์š”์†Œ ํƒ€์ž…์ด ๋ฉ”ํƒ€๋ฐ์ดํ„ฐ์— ์ €์žฅ๋ฉ๋‹ˆ๋‹ค. ํ—ค๋”๋ฅผ ์ฒญํฌ ๋ฉ”ํƒ€๋ฐ์ดํ„ฐ๋กœ ํ™œ์šฉํ•˜๊ฑฐ๋‚˜ ํ…Œ์ด๋ธ”๋งŒ ๋”ฐ๋กœ ์ถ”์ถœํ•  ๋•Œ ์”๋‹ˆ๋‹ค.",
      "**API vs ๋กœ์ปฌ**: Unstructured Cloud API๋Š” ๊ณ ํ’ˆ์งˆ OCR๊ณผ ๋ ˆ์ด์•„์›ƒ ๋ถ„์„์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ๋กœ์ปฌ์€ ๋ฌด๋ฃŒ์ด๋‚˜ ์ด๋ฏธ์ง€๋‚˜ ๋ณต์žกํ•œ ๋ ˆ์ด์•„์›ƒ์˜ PDF ํ’ˆ์งˆ์ด ๋‚ฎ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค."
    ]
  },
  {
    "id": "docs-conversation-memory",
    "certification": "Docs Study",
    "title": "LangChain ๋Œ€ํ™” ๋ฉ”๋ชจ๋ฆฌ โ€” ๋ฉ€ํ‹ฐํ„ด ์ฑ—๋ด‡",
    "level": "๊ณ ๊ธ‰",
    "related_practices": ["docs-lab-6", "docs-brain-1"],
    "summary": "RunnableWithMessageHistory๋Š” LangChain ์ฒด์ธ์— ๋Œ€ํ™” ํžˆ์Šคํ† ๋ฆฌ๋ฅผ ์ž๋™์œผ๋กœ ์ฃผ์ž…ํ•ฉ๋‹ˆ๋‹ค. ๋งค ์š”์ฒญ๋งˆ๋‹ค ์ด์ „ ๋Œ€ํ™”๋ฅผ ์ง์ ‘ ๊ด€๋ฆฌํ•˜์ง€ ์•Š์•„๋„ session_id๋งŒ ๋„˜๊ธฐ๋ฉด ํžˆ์Šคํ† ๋ฆฌ๊ฐ€ ์œ ์ง€๋ฉ๋‹ˆ๋‹ค.",
    "example": "from langchain_community.chat_message_histories import ChatMessageHistory\nfrom langchain_core.runnables.history import RunnableWithMessageHistory\n\nstore = {}  # session_id โ†’ ChatMessageHistory\n\ndef get_session_history(session_id: str):\n    if session_id not in store:\n        store[session_id] = ChatMessageHistory()\n    return store[session_id]\n\nchain_with_history = RunnableWithMessageHistory(\n    chain,\n    get_session_history,\n    input_messages_key=\"question\",\n    history_messages_key=\"chat_history\"\n)\n\nchain_with_history.invoke(\n    {\"question\": \"์•ˆ๋…•\"},\n    config={\"configurable\": {\"session_id\": \"user-1\"}}\n)",
    "common_mistake": "`session_id`๋ฅผ ๋ชจ๋“  ์‚ฌ์šฉ์ž์—๊ฒŒ ๋™์ผํ•˜๊ฒŒ ๋„˜๊ธฐ๋ฉด ์ „์ฒด ์‚ฌ์šฉ์ž๊ฐ€ ํ•˜๋‚˜์˜ ๋Œ€ํ™” ํžˆ์Šคํ† ๋ฆฌ๋ฅผ ๊ณต์œ ํ•ฉ๋‹ˆ๋‹ค. ์‚ฌ์šฉ์ž๋ณ„๋กœ ๊ณ ์œ ํ•œ session_id๋ฅผ ์ƒ์„ฑํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
    "keywords": ["RunnableWithMessageHistory", "ChatMessageHistory", "session_id", "chat_history", "๋ฉ€ํ‹ฐํ„ด"],
    "source_id": "official-docs-langchain",
    "details": [
      "**ํžˆ์Šคํ† ๋ฆฌ ์ €์žฅ์†Œ**: ์ธ๋ฉ”๋ชจ๋ฆฌ(`ChatMessageHistory`)๋Š” ์„œ๋ฒ„ ์žฌ์‹œ์ž‘ ์‹œ ์‚ฌ๋ผ์ง‘๋‹ˆ๋‹ค. ํ”„๋กœ๋•์…˜์—์„œ๋Š” `RedisChatMessageHistory`, `SQLChatMessageHistory` ๋“ฑ ์˜์† ์ €์žฅ์†Œ๋ฅผ ์”๋‹ˆ๋‹ค.",
      "**ํžˆ์Šคํ† ๋ฆฌ ๊ธธ์ด ์ œํ•œ**: ๋Œ€ํ™”๊ฐ€ ๊ธธ์–ด์ง€๋ฉด ํ† ํฐ์ด ์ดˆ๊ณผ๋ฉ๋‹ˆ๋‹ค. `trim_messages()` ๋˜๋Š” ์ตœ๊ทผ N๊ฐœ ๋ฉ”์‹œ์ง€๋งŒ ์œ ์ง€ํ•˜๋Š” ๋กœ์ง์„ ์ถ”๊ฐ€ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.",
      "**ํ”„๋กฌํ”„ํŠธ ์—ฐ๋™**: `ChatPromptTemplate`์— `MessagesPlaceholder(variable_name=\"chat_history\")`๋ฅผ ์ถ”๊ฐ€ํ•ด์•ผ ํžˆ์Šคํ† ๋ฆฌ๊ฐ€ ํ”„๋กฌํ”„ํŠธ์— ์ฃผ์ž…๋ฉ๋‹ˆ๋‹ค. history_messages_key์™€ placeholder ์ด๋ฆ„์ด ์ผ์น˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค."
    ]
  }
]