initial

main.py

@@ -0,0 +1,20 @@
+from typing import Any, Optional
+
+from fastapi import FastAPI
+from pydantic import BaseModel
+
+app = FastAPI()
+
+class APIResponse(BaseModel):
+    message: str
+    statusCode: int
+    payload: Optional[Any] = None
+    
+# python -m uvicorn main:app --reload
+@app.get("/")
+def home():
+    return APIResponse(
+        message="Job Processor API is running",
+        statusCode=200,
+        payload=None
+    )

services/cv_pipeline.py

@@ -0,0 +1,123 @@
+"""
+pip install pdfplumber python-docx transformers torch accelerate
+"""
+
+import io
+import re
+import json
+import pdfplumber
+from docx import Document
+from docx.oxml.ns import qn
+from transformers import pipeline
+
+
+# ── local llm ────────────────────────────────────────────────────────────────
+# Qwen2.5-1.5B-Instruct: ~3GB RAM, follows JSON instructions reliably
+# Swap model= for anything larger if you have more RAM:
+#   "Qwen/Qwen2.5-3B-Instruct"   (~6GB)
+#   "Qwen/Qwen2.5-7B-Instruct"   (~14GB)
+
+llm = pipeline(
+    "text-generation",
+    model="Qwen/Qwen2.5-1.5B-Instruct",
+    device_map="auto",       # GPU if available, else CPU
+    torch_dtype="auto",
+)
+
+SYSTEM_PROMPT = """You are a CV parser. Extract information and return ONLY raw JSON — no markdown, no backticks, no explanation.
+
+Schema (null for missing, [] for empty arrays):
+{
+  "contact": { "full_name": str|null, "email": str|null, "phone": str|null, "location": str|null, "linkedin": str|null, "github": str|null },
+  "summary": str|null,
+  "experience": [{ "company": str|null, "title": str|null, "start_date": str|null, "end_date": str|null, "description": [str] }],
+  "education":  [{ "institution": str|null, "degree": str|null, "field_of_study": str|null, "start_date": str|null, "end_date": str|null }],
+  "skills": [str],
+  "certifications": [{ "name": str|null, "issuer": str|null, "date": str|null }],
+  "languages": [str]
+}"""
+
+
+# ── extraction ────────────────────────────────────────────────────────────────
+
+def extract(file_bytes: bytes) -> str:
+    if file_bytes[:4] == b"%PDF":
+        return _from_pdf(file_bytes)
+    if file_bytes[:2] == b"PK":
+        return _from_docx(file_bytes)
+    raise ValueError("Unsupported file type. Upload PDF or DOCX.")
+
+
+def _from_pdf(data: bytes) -> str:
+    parts = []
+    with pdfplumber.open(io.BytesIO(data)) as pdf:
+        for page in pdf.pages:
+            t = page.extract_text(x_tolerance=2, y_tolerance=2)
+            if t:
+                parts.append(t)
+    return "\n".join(parts)
+
+
+def _from_docx(data: bytes) -> str:
+    doc = Document(io.BytesIO(data))
+    parts = []
+    for child in doc.element.body:
+        tag = child.tag.split("}")[-1]
+        if tag == "p":
+            parts.append("".join(n.text or "" for n in child.iter(qn("w:t"))))
+        elif tag == "tbl":
+            for row in child.iter(qn("w:tr")):
+                cells = ["".join(n.text or "" for n in cell.iter(qn("w:t"))) for cell in row.iter(qn("w:tc"))]
+                parts.append("\t".join(cells))
+    return "\n".join(p for p in parts if p.strip())
+
+
+# ── clean ─────────────────────────────────────────────────────────────────────
+
+def clean(text: str) -> str:
+    text = re.sub(r"-\s*\n\s*", "", text)
+    text = re.sub(r"\n{3,}", "\n\n", text)
+    text = re.sub(r"[\x00-\x08\x0b\x0c\x0e-\x1f]", "", text)
+    text = text.replace("\u00a0", " ").replace("\u2013", "-").replace("\u2014", "-")
+    return text.strip()
+
+
+# ── llm ───────────────────────────────────────────────────────────────────────
+
+def call_llm(text: str) -> dict:
+    messages = [
+        {"role": "system",  "content": SYSTEM_PROMPT},
+        {"role": "user",    "content": text},
+    ]
+
+    out = llm(
+        messages,
+        max_new_tokens=2000,
+        do_sample=False,        # greedy = deterministic JSON
+        temperature=None,       # must be None when do_sample=False
+        top_p=None,
+    )
+
+    raw = out[0]["generated_text"][-1]["content"]   # last assistant turn
+    raw = re.sub(r"^```(?:json)?\s*|\s*```$", "", raw.strip())
+    return json.loads(raw)
+
+
+# ── pipeline ──────────────────────────────────────────────────────────────────
+
+def process(file_bytes: bytes) -> dict:
+    text = extract(file_bytes)
+    if not text.strip():
+        raise ValueError("No text extracted from document.")
+    return call_llm(clean(text))
+
+
+# ── cli ───────────────────────────────────────────────────────────────────────
+
+if __name__ == "__main__":
+    import sys
+    from pathlib import Path
+
+    path = Path(sys.argv[1])
+    result = process(path.read_bytes())
+    print(json.dumps(result, indent=2))