Spaces:
Paused
Paused
File size: 20,335 Bytes
a01e1da 0a031ed a01e1da f72c466 a01e1da f9c397a a01e1da f9c397a a01e1da 9a8f2bd a01e1da 9a8f2bd a01e1da 9a8f2bd a01e1da | 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 | """
LexMind — FastAPI Backend (Pinecone + HuggingFace Inference API)
Run with: uvicorn main:app --reload --port 8000
"""
import os
import re
import json
from pathlib import Path
from typing import Optional
import httpx
import fitz # PyMuPDF
import torch
from fastapi import FastAPI, UploadFile, File, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from fastapi.staticfiles import StaticFiles
from fastapi.responses import FileResponse
from pydantic import BaseModel
from sentence_transformers import SentenceTransformer
from pinecone import Pinecone
from dotenv import load_dotenv
load_dotenv()
# ── Configuration ─────────────────────────────────────────────────────────────
PINECONE_API_KEY = os.getenv("pinecone", "")
HF_API_KEY = os.getenv("hf_tokens", "")
JUDGEMENTS_INDEX = "legal-judgements"
LEGAL_FRAMEWORK_INDEX = "legal-framework"
LOCAL_MODEL_DIR = "./models/bge-small"
EMBED_MODEL_NAME = "BAAI/bge-small-en-v1.5"
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
# Both stages use the same model — change here to use different ones
HF_ROUTER_MODEL = "meta-llama/Llama-3.1-8B-Instruct" # Stage 1: conversation + routing
HF_LEGAL_MODEL = "meta-llama/Llama-3.1-8B-Instruct" # Stage 2: legal RAG answer
HF_CHAT_URL = "https://router.huggingface.co/v1/chat/completions"
BGE_PREFIX = "Represent this sentence for searching relevant passages: "
TOP_K = 10
CONSTITUTION_TOP_K = 5
# ─────────────────────────────────────────────────────────────────────────────
# ── Load embedding model ──────────────────────────────────────────────────────
def load_embed_model() -> SentenceTransformer:
local = Path(LOCAL_MODEL_DIR)
if local.exists() and any(local.iterdir()):
print(f"✅ Loading bge-small from '{LOCAL_MODEL_DIR}'")
else:
print(f"📥 Downloading {EMBED_MODEL_NAME} (~130 MB)…")
local.mkdir(parents=True, exist_ok=True)
m = SentenceTransformer(EMBED_MODEL_NAME)
m.save(str(local))
print(f"✅ Model saved to '{LOCAL_MODEL_DIR}'")
model = SentenceTransformer(str(local))
model = model.to(DEVICE)
print(f" Embedding device: {DEVICE}")
return model
embed_model = load_embed_model()
# ── Connect to Pinecone ───────────────────────────────────────────────────────
print("🔌 Connecting to Pinecone…")
pc = Pinecone(api_key=PINECONE_API_KEY)
judgements_index = pc.Index(JUDGEMENTS_INDEX)
print(f"✅ Judgements index | vectors: {judgements_index.describe_index_stats().total_vector_count}")
try:
legal_index = pc.Index(LEGAL_FRAMEWORK_INDEX)
print(f"✅ Legal framework index | vectors: {legal_index.describe_index_stats().total_vector_count}")
except Exception:
legal_index = None
print("⚠️ Legal framework index not found — run build_pinecone_legal.py.")
# ── FastAPI app ───────────────────────────────────────────────────────────────
app = FastAPI(title="LexMind API", version="3.0.0")
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_methods=["*"],
allow_headers=["*"],
)
# ── Pydantic models ───────────────────────────────────────────────────────────
class SearchRequest(BaseModel):
query: str
top_k: int = 10
offset: int = 0
year_from: Optional[int] = None
year_to: Optional[int] = None
class ChatRequest(BaseModel):
message: str
context: str = ""
system_prompt: str = ""
model_override: str = ""
class DroppedCitationModel(BaseModel):
file_name: str = ""
year: str = ""
content: str = ""
score: float = 0.0
class SmartChatRequest(BaseModel):
message: str
case_text: str = "" # user's case description
dropped_citation: Optional[DroppedCitationModel] = None # only if user dragged a doc
# ── HuggingFace helper ────────────────────────────────────────────────────────
async def call_hf(
model: str,
system: str,
user: str,
temperature: float = 0.4,
max_tokens: int = 1024,
timeout: int = 120,
) -> str:
headers = {
"Authorization": f"Bearer {HF_API_KEY}",
"Content-Type": "application/json",
}
payload = {
"model": model,
"messages": [
{"role": "system", "content": system},
{"role": "user", "content": user},
],
"max_tokens": max_tokens,
"temperature": temperature,
"top_p": 0.9,
"stream": False,
}
async with httpx.AsyncClient(timeout=timeout) as client:
r = await client.post(HF_CHAT_URL, headers=headers, json=payload)
if r.status_code != 200:
print(f"[HF ERROR] status={r.status_code} model={model} body={r.text[:400]}")
if r.status_code == 401:
raise HTTPException(status_code=401,
detail="Invalid HuggingFace API key.")
if r.status_code == 403:
raise HTTPException(status_code=403,
detail=f"Access denied for '{model}'. Accept the license at huggingface.co/{model}")
if r.status_code == 404:
raise HTTPException(status_code=404,
detail=f"Model '{model}' not found.")
if r.status_code == 429:
raise HTTPException(status_code=429,
detail="HuggingFace rate limit hit. Please wait and retry.")
if r.status_code == 503:
raise HTTPException(status_code=503,
detail=f"Model '{model}' is loading (~20s). Please retry.")
r.raise_for_status()
data = r.json()
choices = data.get("choices", [])
if choices:
content = choices[0].get("message", {}).get("content", "")
if content:
return content.strip()
if isinstance(data, list) and data:
return data[0].get("generated_text", "").strip()
raise HTTPException(status_code=500,
detail=f"Unexpected HF response: {str(data)[:200]}")
# ── Embed helper ──────────────────────────────────────────────────────────────
def embed_query(text: str) -> list[float]:
return embed_model.encode(
BGE_PREFIX + text,
normalize_embeddings=True,
device=DEVICE
).tolist()
# ── Routes ────────────────────────────────────────────────────────────────────
@app.get("/api/health")
async def health():
hf_ok = False
try:
async with httpx.AsyncClient(timeout=5) as client:
r = await client.get(
"https://huggingface.co/api/whoami",
headers={"Authorization": f"Bearer {HF_API_KEY}"}
)
hf_ok = r.status_code == 200
except Exception:
pass
j_stats = judgements_index.describe_index_stats()
l_stats = legal_index.describe_index_stats() if legal_index else None
return {
"status": "ok",
"huggingface": "authenticated" if hf_ok else "check HF_API_KEY",
"router_model": HF_ROUTER_MODEL,
"legal_model": HF_LEGAL_MODEL,
"judgements_vectors": j_stats.total_vector_count,
"legal_vectors": l_stats.total_vector_count if l_stats else 0,
"embed_device": DEVICE,
}
@app.post("/api/search")
async def search(req: SearchRequest):
"""Semantic search over judgements Pinecone index with pagination and optional year filtering."""
if not req.query.strip():
raise HTTPException(status_code=400, detail="Query cannot be empty.")
has_year_filter = req.year_from is not None and req.year_to is not None
if has_year_filter:
fetch_k = min(300, max(req.offset + req.top_k * 10, 150))
else:
fetch_k = min(req.offset + req.top_k, 100)
try:
result = judgements_index.query(
vector=embed_query(req.query),
top_k=fetch_k,
include_metadata=True,
)
except Exception as e:
raise HTTPException(status_code=500, detail=f"Search failed: {str(e)}")
output = []
for m in result.get("matches", []):
meta = m.get("metadata", {})
year_str = meta.get("year", "unknown")
if has_year_filter:
try:
year_int = int(year_str)
if not (req.year_from <= year_int <= req.year_to):
continue
except (ValueError, TypeError):
continue
output.append({
"file_name": meta.get("file_name", "Unknown"),
"year": year_str,
"source": meta.get("source", ""),
"score": round(float(m.get("score", 0)), 4),
"content": meta.get("content", ""),
})
output.sort(key=lambda x: x["score"], reverse=True)
paginated = output[req.offset: req.offset + req.top_k]
return {
"results": paginated,
"count": len(output),
}
@app.post("/api/extract-pdf")
async def extract_pdf(file: UploadFile = File(...)):
"""Extract full text from an uploaded PDF."""
if not file.filename.lower().endswith(".pdf"):
raise HTTPException(status_code=400, detail="Only PDF files are accepted.")
contents = await file.read()
try:
doc = fitz.open(stream=contents, filetype="pdf")
pages = [page.get_text() for page in doc]
doc.close()
text = "\n\n".join(pages).strip()
except Exception as e:
raise HTTPException(status_code=500, detail=f"PDF extraction failed: {str(e)}")
return {"text": text, "pages": len(pages), "filename": file.filename}
@app.post("/api/legal-context")
async def legal_context(req: SearchRequest):
"""Retrieve legal framework chunks from Pinecone."""
if not legal_index:
return {"results": [], "count": 0}
if not req.query.strip():
raise HTTPException(status_code=400, detail="Query cannot be empty.")
try:
result = legal_index.query(
vector=embed_query(req.query),
top_k=min(req.top_k or CONSTITUTION_TOP_K, 10),
include_metadata=True,
)
except Exception as e:
raise HTTPException(status_code=500, detail=f"Legal context search failed: {str(e)}")
output = []
for m in result.get("matches", []):
meta = m.get("metadata", {})
output.append({
"source": meta.get("source", "Unknown"),
"type": meta.get("type", ""),
"section": meta.get("section", ""),
"score": round(float(m.get("score", 0)), 4),
"content": meta.get("content", ""),
})
output.sort(key=lambda x: x["score"], reverse=True)
return {"results": output, "count": len(output)}
@app.post("/api/chat")
async def chat_legacy(req: ChatRequest):
"""Legacy endpoint — used by CitationCard summarize and AI compare features."""
system = (
"You are LexMind, a professional Indian legal research assistant. "
"Answer concisely and professionally based only on the provided context."
)
user = (
f"CONTEXT:\n{req.context}\n\nQUESTION: {req.message}"
if req.context.strip() else req.message
)
try:
reply = await call_hf(HF_LEGAL_MODEL, system, user)
return {"reply": reply}
except HTTPException:
raise
except Exception as e:
raise HTTPException(status_code=500, detail=f"Chat failed: {str(e)}")
@app.post("/api/smart-chat")
async def smart_chat(req: SmartChatRequest):
"""
Two-stage conversational chat:
Stage 1 — LLM1 (Llama-3.1-8B):
- Always knows the user's case description
- Handles casual conversation naturally
- If legal question detected, produces a precise rag_query for LLM2
- Has NO knowledge of retrieved judgements
- Only knows about a dropped citation if user explicitly dragged one in
Stage 2 — LLM2 (Llama-3.1-8B):
- Only called when Stage 1 detects a legal question
- Gets: legal framework from Pinecone + dropped citation (if any)
- Returns grounded legal answer with [LAW: source] citations
"""
# ── Build case context for LLM1 ──────────────────────────────────────────
case_ctx = ""
if req.case_text.strip():
case_ctx = f"\nCURRENT USER CASE:\n{req.case_text[:800]}\n"
dropped_ctx = ""
if req.dropped_citation and req.dropped_citation.content.strip():
name = (req.dropped_citation.file_name or '').replace('_', ' ').strip()
dropped_ctx = (
f"\nUSER HAS SHARED THIS JUDGEMENT FOR DISCUSSION:\n"
f"Case: {name} ({req.dropped_citation.year or '?'})\n"
f"{req.dropped_citation.content[:2000]}\n"
)
# ── Stage 1: Router + conversationalist ──────────────────────────────────
router_system = f"""You are LexMind, a friendly and professional Indian legal research assistant.
{case_ctx}{dropped_ctx}
YOUR BEHAVIOUR:
- For casual messages (greetings, thanks, small talk): reply naturally and warmly in 1-2 sentences.
- For questions about the shared judgement above (if any): you can answer directly from it.
- For legal questions requiring Constitution/IPC/CrPC/BSA knowledge: identify what needs to be looked up.
- Never make up legal information you are not sure about.
Respond ONLY with valid JSON, no extra text, no markdown fences:
For casual chat:
{{"intent": "chat", "response": "your warm friendly reply here", "rag_query": null}}
For a legal question you can answer from the shared judgement:
{{"intent": "citation", "response": "your answer from the judgement", "rag_query": null}}
For a legal question needing Constitution/IPC/CrPC/BSA lookup:
{{"intent": "legal", "response": null, "rag_query": "precise 3-8 word search query"}}"""
router_user = f'User message: "{req.message}"'
try:
raw = await call_hf(
HF_ROUTER_MODEL,
router_system,
router_user,
temperature=0.2,
max_tokens=300,
timeout=60,
)
except HTTPException:
raise
except Exception as e:
raise HTTPException(status_code=500, detail=f"Stage 1 failed: {str(e)}")
# ── Parse Stage 1 JSON ────────────────────────────────────────────────────
intent = "chat"
response = None
rag_query = None
try:
clean = re.sub(r"```json|```", "", raw).strip()
match = re.search(r"\{.*\}", clean, re.DOTALL)
parsed = json.loads(match.group(0) if match else clean)
intent = parsed.get("intent", "chat")
response = parsed.get("response")
rag_query = parsed.get("rag_query")
except Exception:
# JSON parse failed — treat raw text as a casual reply
intent = "chat"
response = raw.strip() if raw.strip() else "How can I help you?"
# ── Stage 1 exits: casual or citation answer ──────────────────────────────
if intent in ("chat", "citation"):
return {
"reply": response or "How can I help you today?",
"intent": intent,
}
# ── Stage 2: Legal RAG answer ─────────────────────────────────────────────
search_q = rag_query or req.message
# 2a. Search Pinecone legal-framework index
legal_ctx = ""
if legal_index and search_q:
try:
law_result = legal_index.query(
vector=embed_query(search_q),
top_k=CONSTITUTION_TOP_K,
include_metadata=True,
)
matches = sorted(
law_result.get("matches", []),
key=lambda x: x.get("score", 0),
reverse=True,
)
if matches:
legal_ctx = "RELEVANT LEGAL FRAMEWORK (Constitution / IPC / CrPC / BSA):\n\n"
for m in matches:
meta = m.get("metadata", {})
src = meta.get("source", "Law")
sec = meta.get("section", "")
legal_ctx += f"[LAW: {src}{' S.' + str(sec) if sec else ''}]\n"
legal_ctx += f"{meta.get('content', '')[:600]}\n\n---\n\n"
except Exception:
pass # continue without legal context
# 2b. Build Stage 2 context
# Includes: case description + dropped citation (if any) + legal framework
# Does NOT include retrieved judgements
stage2_context = ""
if req.case_text.strip():
stage2_context += f"USER'S CASE:\n{req.case_text[:800]}\n\n"
if dropped_ctx:
stage2_context += dropped_ctx + "\n"
if legal_ctx:
stage2_context += legal_ctx
legal_system = """You are LexMind, a professional Indian legal research assistant.
KNOWLEDGE BASE YOU CAN USE:
- The user's case description (if provided)
- A shared judgement (if user dragged one in)
- Indian Constitution, IPC, CrPC, BSA 2023 — cited as [LAW: source S.section]
KNOWLEDGE GAPS — be honest if asked about these:
- Code of Civil Procedure (CPC) — not in your knowledge base
- Indian Contract Act — not in your knowledge base
- Transfer of Property Act — not in your knowledge base
RULES:
1. Answer ONLY from the provided context. Never fabricate.
2. Cite laws as [LAW: IPC S.302] or [LAW: Indian Constitution Art.21].
3. If context is insufficient: "I don't have enough information on this. Please search for relevant citations."
4. Be concise, clear, and professional.
5. Answer directly — no preamble like "Based on the context provided…"."""
legal_user = (
f"QUESTION: {req.message}\n\nCONTEXT:\n{stage2_context}"
if stage2_context.strip()
else req.message
)
try:
reply = await call_hf(
HF_LEGAL_MODEL,
legal_system,
legal_user,
temperature=0.2,
max_tokens=1024,
timeout=120,
)
return {"reply": reply, "intent": "legal"}
except HTTPException:
raise
except Exception as e:
raise HTTPException(status_code=500, detail=f"Stage 2 failed: {str(e)}")
# ── Serve React frontend ──────────────────────────────────────────────────────
# Built frontend output is generated under ../frontend/dist (relative to backend/)
dist_path = Path("../frontend/dist")
if dist_path.exists():
app.mount("/assets", StaticFiles(directory=str(dist_path / "assets")), name="assets")
@app.get("/")
async def serve_frontend():
return FileResponse(str(dist_path / "index.html"))
@app.get("/{full_path:path}")
async def serve_spa(full_path: str):
return FileResponse(str(dist_path / "index.html"))
|