Spaces:
Running
Running
Add RAG Document Q&A app
Browse files- README.md +15 -6
- index.html +98 -19
README.md
CHANGED
|
@@ -1,10 +1,19 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
-
colorFrom:
|
| 5 |
-
colorTo:
|
| 6 |
sdk: static
|
| 7 |
-
pinned:
|
|
|
|
|
|
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
title: RAG Document Q&A
|
| 3 |
+
emoji: 📚
|
| 4 |
+
colorFrom: blue
|
| 5 |
+
colorTo: indigo
|
| 6 |
sdk: static
|
| 7 |
+
pinned: true
|
| 8 |
+
license: mit
|
| 9 |
+
short_description: "In-browser RAG: ask questions about your own PDFs"
|
| 10 |
+
tags: [rag, retrieval, question-answering, transformers.js, semantic-search]
|
| 11 |
---
|
| 12 |
|
| 13 |
+
# 📚 RAG Document Q&A
|
| 14 |
+
|
| 15 |
+
Retrieval-Augmented Generation pipeline that runs entirely in the browser: document loading (PDF via pdf.js, TXT, pasted text), chunking with overlap, sentence embeddings (Xenova/all-MiniLM-L6-v2), cosine-similarity vector search and extractive answering (Xenova/distilbert-base-cased-distilled-squad). Retrieved passages are shown with similarity scores so every answer is traceable. Documents never leave the user's device.
|
| 16 |
+
|
| 17 |
+
Runs fully client-side with [Transformers.js](https://huggingface.co/docs/transformers.js) (ONNX Runtime Web) — no server, no API keys.
|
| 18 |
+
|
| 19 |
+
**Author:** [Alexandru Cojocari](https://www.linkedin.com/in/alexdataanalyst/) · [GitHub](https://github.com/DatadrivenAlex) — Junior AI Developer (Generative AI, RAG, Agentic AI, Secure AI)
|
index.html
CHANGED
|
@@ -1,19 +1,98 @@
|
|
| 1 |
-
<!
|
| 2 |
-
<html>
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
<
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<!DOCTYPE html>
|
| 2 |
+
<html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
|
| 3 |
+
<title>RAG Document Q&A</title><link rel="stylesheet" href="style.css"></head>
|
| 4 |
+
<body><main>
|
| 5 |
+
<h1>📚 RAG Document Q&A</h1>
|
| 6 |
+
<p class="sub">Retrieval-Augmented question answering over your own documents, running 100% in your browser.
|
| 7 |
+
Pipeline: load (PDF/TXT/paste) → chunk with overlap → embed (all-MiniLM-L6-v2) → cosine vector search → extractive answer (DistilBERT-SQuAD), with sources.</p>
|
| 8 |
+
|
| 9 |
+
<div class="card">
|
| 10 |
+
<div class="row">
|
| 11 |
+
<div><label>Documents (PDF, TXT, MD)</label><input id="files" type="file" multiple accept=".pdf,.txt,.md">
|
| 12 |
+
<p class="muted">Files never leave your device.</p></div>
|
| 13 |
+
<div><label>…or paste text</label><textarea id="pasted"></textarea></div>
|
| 14 |
+
</div>
|
| 15 |
+
<button id="build">1. Build vector index</button>
|
| 16 |
+
<p id="status" class="muted"></p><progress id="prog" value="0" max="100" hidden></progress>
|
| 17 |
+
</div>
|
| 18 |
+
|
| 19 |
+
<div class="card">
|
| 20 |
+
<label>2. Ask a question</label><input id="q" type="text" value="Why does RAG reduce hallucinations?">
|
| 21 |
+
<label>Passages to retrieve (top-k): <span id="kv">3</span></label><input id="k" type="range" min="1" max="8" value="3">
|
| 22 |
+
<br><button id="ask" disabled>Ask</button>
|
| 23 |
+
<div id="answer"></div><div id="sources"></div>
|
| 24 |
+
</div>
|
| 25 |
+
|
| 26 |
+
<footer>Stack: Transformers.js · ONNX Runtime Web · pdf.js · vanilla JS. Built by
|
| 27 |
+
<a href="https://www.linkedin.com/in/alexdataanalyst/" target="_blank">Alexandru Cojocari</a> ·
|
| 28 |
+
<a href="https://github.com/DatadrivenAlex" target="_blank">GitHub</a></footer>
|
| 29 |
+
</main>
|
| 30 |
+
<script type="module">
|
| 31 |
+
import { pipeline } from 'https://cdn.jsdelivr.net/npm/@huggingface/transformers@3';
|
| 32 |
+
import * as pdfjs from 'https://cdn.jsdelivr.net/npm/pdfjs-dist@4.10.38/build/pdf.min.mjs';
|
| 33 |
+
pdfjs.GlobalWorkerOptions.workerSrc = 'https://cdn.jsdelivr.net/npm/pdfjs-dist@4.10.38/build/pdf.worker.min.mjs';
|
| 34 |
+
|
| 35 |
+
const $ = id => document.getElementById(id);
|
| 36 |
+
$('pasted').value = `Retrieval-Augmented Generation (RAG) combines a retriever and a generator.
|
| 37 |
+
Documents are split into chunks, converted into embeddings and stored in a vector database such as FAISS or ChromaDB.
|
| 38 |
+
At query time the question is embedded, the most similar chunks are retrieved and passed to a language model as context.
|
| 39 |
+
RAG reduces hallucinations because answers are grounded in retrieved sources, and it lets a model use private or recent data without fine-tuning.
|
| 40 |
+
Typical frameworks for building RAG pipelines are LangChain, LlamaIndex and LangGraph for agentic workflows.`;
|
| 41 |
+
$('k').oninput = e => $('kv').textContent = e.target.value;
|
| 42 |
+
|
| 43 |
+
let embedder, qa, index = null;
|
| 44 |
+
const progress = info => { if (info.status === 'progress') { $('prog').hidden = false; $('prog').value = info.progress; } };
|
| 45 |
+
const esc = s => s.replace(/[&<>]/g, c => ({'&':'&','<':'<','>':'>'}[c]));
|
| 46 |
+
|
| 47 |
+
async function readFile(f) {
|
| 48 |
+
if (f.name.toLowerCase().endsWith('.pdf')) {
|
| 49 |
+
const pdf = await pdfjs.getDocument({ data: await f.arrayBuffer() }).promise;
|
| 50 |
+
let out = '';
|
| 51 |
+
for (let i = 1; i <= pdf.numPages; i++) out += (await (await pdf.getPage(i)).getTextContent()).items.map(t => t.str).join(' ') + '\n';
|
| 52 |
+
return out;
|
| 53 |
+
}
|
| 54 |
+
return await f.text();
|
| 55 |
+
}
|
| 56 |
+
function chunk(text, size = 500, overlap = 100) {
|
| 57 |
+
text = text.replace(/\s+/g, ' ').trim(); const out = [];
|
| 58 |
+
for (let s = 0; s < text.length; s += size - overlap) out.push(text.slice(s, s + size));
|
| 59 |
+
return out.filter(c => c.length > 30);
|
| 60 |
+
}
|
| 61 |
+
const dot = (a, b) => a.reduce((s, v, i) => s + v * b[i], 0);
|
| 62 |
+
|
| 63 |
+
$('build').onclick = async () => {
|
| 64 |
+
$('build').disabled = true;
|
| 65 |
+
try {
|
| 66 |
+
let text = $('pasted').value;
|
| 67 |
+
for (const f of $('files').files) text += '\n\n' + await readFile(f);
|
| 68 |
+
const chunks = chunk(text);
|
| 69 |
+
if (!chunks.length) { $('status').textContent = 'Please add a document or text.'; return; }
|
| 70 |
+
$('status').textContent = 'Loading embedding model…';
|
| 71 |
+
embedder ??= await pipeline('feature-extraction', 'Xenova/all-MiniLM-L6-v2', { progress_callback: progress });
|
| 72 |
+
$('status').textContent = `Embedding ${chunks.length} chunks…`;
|
| 73 |
+
const vecs = (await embedder(chunks, { pooling: 'mean', normalize: true })).tolist();
|
| 74 |
+
index = chunks.map((c, i) => ({ c, v: vecs[i] }));
|
| 75 |
+
$('status').textContent = 'Loading QA model…';
|
| 76 |
+
qa ??= await pipeline('question-answering', 'Xenova/distilbert-base-cased-distilled-squad', { progress_callback: progress });
|
| 77 |
+
$('prog').hidden = true;
|
| 78 |
+
$('status').innerHTML = `<span class="ok">✔ Indexed ${chunks.length} chunks.</span> Ask a question below.`;
|
| 79 |
+
$('ask').disabled = false;
|
| 80 |
+
} catch (e) { $('status').innerHTML = `<span class="bad">Error: ${esc(e.message)}</span>`; }
|
| 81 |
+
finally { $('build').disabled = false; }
|
| 82 |
+
};
|
| 83 |
+
|
| 84 |
+
$('ask').onclick = async () => {
|
| 85 |
+
const q = $('q').value.trim(); if (!q || !index) return;
|
| 86 |
+
$('ask').disabled = true; $('answer').textContent = 'Thinking…';
|
| 87 |
+
try {
|
| 88 |
+
const qv = (await embedder(q, { pooling: 'mean', normalize: true })).tolist()[0];
|
| 89 |
+
const hits = index.map(x => ({ ...x, s: dot(qv, x.v) })).sort((a, b) => b.s - a.s).slice(0, +$('k').value);
|
| 90 |
+
const res = await qa(q, hits.map(h => h.c).join(' '));
|
| 91 |
+
$('answer').innerHTML = `<h3>Answer</h3><p style="font-size:18px"><b>${esc(res.answer || '—')}</b>
|
| 92 |
+
<span class="muted">(confidence ${(res.score ?? 0).toFixed(2)})</span></p>`;
|
| 93 |
+
$('sources').innerHTML = '<h3>Sources</h3>' + hits.map((h, i) =>
|
| 94 |
+
`<p class="muted">[${i + 1}] cosine similarity ${h.s.toFixed(2)}</p><blockquote>${esc(h.c)}</blockquote>`).join('');
|
| 95 |
+
} catch (e) { $('answer').innerHTML = `<span class="bad">Error: ${esc(e.message)}</span>`; }
|
| 96 |
+
finally { $('ask').disabled = false; }
|
| 97 |
+
};
|
| 98 |
+
</script></body></html>
|