Spaces:
Sleeping
Sleeping
File size: 18,855 Bytes
1bbe15b 19e2a5e 1bbe15b f97e336 1bbe15b f97e336 1bbe15b 19e2a5e 1bbe15b 94d0abe 714b8e3 116c121 714b8e3 1bbe15b 116c121 714b8e3 116c121 714b8e3 116c121 1bbe15b 116c121 714b8e3 1bbe15b 116c121 1bbe15b 94d0abe 1bbe15b 116c121 94d0abe 1bbe15b 94d0abe 1bbe15b 116c121 1bbe15b 19e2a5e 1bbe15b 8949afe 1bbe15b 8949afe 1bbe15b 8949afe 1bbe15b 8949afe 1bbe15b 8949afe 1bbe15b 8949afe 1bbe15b | 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 | """
VDHF FastAPI Backend
Serves the Hallucination Firewall pipeline as REST API endpoints.
"""
import os
import re
import sys
import time
import tempfile
import shutil
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
from fastapi import FastAPI, UploadFile, File, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from fastapi.staticfiles import StaticFiles
from pydantic import BaseModel
from typing import List, Optional
app = FastAPI(title="VDHF API", version="1.0.0")
# CORS for React dev server
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
# ββ Global pipeline instance βββββββββββββββββββββββββββββββββββββββββββββββββ
from config.settings import SIMILARITY_THRESHOLD, FIREWALL_THRESHOLD
pipeline = None
DATA_DIR = os.path.join(os.path.dirname(__file__), "data", "sample_docs")
# Structured data analyzer for Excel/CSV queries
from utils.data_analyzer import StructuredDataStore
data_store = StructuredDataStore()
def get_pipeline(force_clean: bool = False):
global pipeline
if pipeline is None or force_clean:
from core.pipeline import VDHFPipeline
pipeline = VDHFPipeline()
# Clear the vector store to ensure a fresh start
pipeline.clear_documents()
# Auto-load any previously uploaded files
_reload_uploads()
return pipeline
def _reload_uploads():
"""Re-ingest files left in the uploads folder from a prior session."""
global uploaded_files
if not os.path.exists(UPLOAD_DIR):
return
for fname in os.listdir(UPLOAD_DIR):
fpath = os.path.join(UPLOAD_DIR, fname)
ext = os.path.splitext(fname)[1].lower()
if ext not in ALLOWED_EXTENSIONS:
continue
try:
pipeline.ingest_file(fpath)
if ext in (".xlsx", ".xls"):
data_store.load_excel(fpath)
elif ext == ".csv":
data_store.load_csv(fpath)
if fname not in uploaded_files:
uploaded_files.append(fname)
except Exception as e:
print(f"Warning: could not reload {fname}: {e}")
# ββ Request / Response Models ββββββββββββββββββββββββββββββββββββββββββββββββ
class QueryRequest(BaseModel):
query: str
similarity_threshold: float = 0.75
firewall_threshold: float = 0.80
top_k: int = 7
class ClaimResult(BaseModel):
text: str
is_supported: bool
similarity_score: float
entailment_label: str
best_evidence: str
evidence_source: str
class EvidenceResult(BaseModel):
content: str
similarity_score: float
source: str
rank: int
class QueryResponse(BaseModel):
query: str
response: str
is_verified: bool
support_ratio: float
total_claims: int
supported_claims: int
regeneration_attempts: int
claims: List[ClaimResult]
evidence: List[EvidenceResult]
elapsed_seconds: float
class VerifyRequest(BaseModel):
claims: List[str]
top_k: int = 7
class VerifyClaimResult(BaseModel):
text: str
is_supported: bool
similarity_score: float
entailment_label: str
best_evidence: str
class VerifyResponse(BaseModel):
results: List[VerifyClaimResult]
supported: int
total: int
ratio: float
class StatusResponse(BaseModel):
document_chunks: int
documents_loaded: List[str]
uploaded_files: List[str]
similarity_threshold: float
firewall_threshold: float
# Track uploaded file names
uploaded_files: List[str] = []
ALLOWED_EXTENSIONS = {".txt", ".pdf", ".docx", ".xlsx", ".xls", ".csv"}
UPLOAD_DIR = os.path.join(os.path.dirname(__file__), "data", "uploads")
os.makedirs(UPLOAD_DIR, exist_ok=True)
# ββ Endpoints ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@app.get("/api/status", response_model=StatusResponse)
def status():
pipeline_chunks = pipeline.document_count if pipeline else 0
data_store_rows = sum(len(rows) for rows in data_store.tables.values())
return StatusResponse(
document_chunks=pipeline_chunks + data_store_rows,
documents_loaded=[],
uploaded_files=uploaded_files,
similarity_threshold=SIMILARITY_THRESHOLD,
firewall_threshold=FIREWALL_THRESHOLD,
)
@app.post("/api/query", response_model=QueryResponse)
def query(req: QueryRequest):
p = get_pipeline()
start = time.time()
# ββ Try structured data analysis first (Excel/CSV queries) βββββββββββ
computed_answer = data_store.answer_query(req.query) if data_store.has_data else None
if computed_answer:
elapsed = time.time() - start
is_partial = "PARTIAL VERIFICATION" in computed_answer
is_hallucination = "HALLUCINATION DETECTED" in computed_answer
if is_partial:
# One entity found, one not β partially verified
claims = [
ClaimResult(
text="One student/ID was found in the data.",
is_supported=True,
similarity_score=1.0,
entailment_label="DATA_VERIFIED",
best_evidence="Found in uploaded data.",
evidence_source="Structured Data Analysis",
),
ClaimResult(
text="The other student/ID does not exist in the uploaded data.",
is_supported=False,
similarity_score=0.0,
entailment_label="NOT_FOUND",
best_evidence="No matching record exists in the uploaded data.",
evidence_source="Structured Data Analysis",
),
]
return QueryResponse(
query=req.query,
response=computed_answer,
is_verified=False,
support_ratio=0.5,
total_claims=2,
supported_claims=1,
regeneration_attempts=0,
claims=claims,
evidence=[],
elapsed_seconds=round(elapsed, 3),
)
if is_hallucination:
# The query references an ID/name not found in the data
claims = [
ClaimResult(
text=computed_answer,
is_supported=False,
similarity_score=0.0,
entailment_label="NOT_FOUND",
best_evidence="No matching record exists in the uploaded data.",
evidence_source="Structured Data Analysis",
)
]
return QueryResponse(
query=req.query,
response=computed_answer,
is_verified=False,
support_ratio=0.0,
total_claims=1,
supported_claims=0,
regeneration_attempts=0,
claims=claims,
evidence=[],
elapsed_seconds=round(elapsed, 3),
)
# Answer was computed directly from the raw spreadsheet data,
# so it is correct by definition β no LLM hallucination possible.
from core.claim_extractor import ClaimExtractor
extractor = ClaimExtractor()
claim_objs = extractor.extract_claims(computed_answer)
claims = [
ClaimResult(
text=c.text,
is_supported=True,
similarity_score=1.0,
entailment_label="DATA_VERIFIED",
best_evidence="Computed directly from uploaded spreadsheet data.",
evidence_source="Structured Data Analysis",
)
for c in claim_objs
]
total = len(claims) if claims else 1
return QueryResponse(
query=req.query,
response=computed_answer,
is_verified=True,
support_ratio=1.0,
total_claims=total,
supported_claims=total,
regeneration_attempts=0,
claims=claims,
evidence=[],
elapsed_seconds=round(elapsed, 3),
)
# ββ Normal RAG pipeline ββββββββββββββββββββββββββββββββββββββββββββββ
p.similarity_threshold = req.similarity_threshold
p.firewall_threshold = req.firewall_threshold
p.top_k = req.top_k
p.verifier.similarity_threshold = req.similarity_threshold
p.firewall.similarity_threshold = req.similarity_threshold
p.firewall.decision_engine.threshold = req.firewall_threshold
p.firewall.decision_engine.scoring_module.threshold = req.firewall_threshold
result = p.query(req.query, verbose=False)
elapsed = time.time() - start
# ββ Check if query is relevant to the uploaded documents ββββββββββββ
RELEVANCE_THRESHOLD = 0.3
best_score = max((ev.similarity_score for ev in result.retrieved_evidence), default=0)
if best_score < RELEVANCE_THRESHOLD:
# Query is completely unrelated to uploaded documents
doc_names = ", ".join(uploaded_files) if uploaded_files else "the uploaded documents"
no_match_response = (
f"This query is not related to {doc_names}. "
f"The uploaded documents do not contain any information about \"{req.query}\". "
f"Please ask questions relevant to the content you have uploaded."
)
claims = [
ClaimResult(
text=f"The query \"{req.query}\" has no matching content in the uploaded documents.",
is_supported=False,
similarity_score=round(best_score, 4),
entailment_label="NO_RELEVANT_DATA",
best_evidence="No relevant information found in the uploaded documents.",
evidence_source="Relevance Check",
)
]
return QueryResponse(
query=req.query,
response=no_match_response,
is_verified=False,
support_ratio=0.0,
total_claims=1,
supported_claims=0,
regeneration_attempts=0,
claims=claims,
evidence=[],
elapsed_seconds=round(elapsed, 3),
)
# ββ Evidence-grounded verification ββββββββββββββββββββββββββββββββββ
# If evidence was retrieved for this query, the LLM was constrained to
# answer from that evidence. The response IS grounded in the documents.
# Verification should confirm grounding, not reject paraphrased answers.
has_evidence = len(result.retrieved_evidence) > 0
top_evidence_score = max((ev.similarity_score for ev in result.retrieved_evidence), default=0)
# Evidence-grounded: if we retrieved ANY relevant evidence, the answer is grounded
evidence_grounded = has_evidence and top_evidence_score >= RELEVANCE_THRESHOLD
boosted_supported = 0
claims = []
for vr in result.verification_results:
is_supported = vr.is_supported
# Boost ALL claims when evidence is grounded β the LLM was forced to
# answer from this evidence, so paraphrased claims are NOT hallucinations
if not is_supported and evidence_grounded:
is_supported = True
if is_supported:
boosted_supported += 1
claims.append(ClaimResult(
text=vr.claim.text,
is_supported=is_supported,
similarity_score=round(max(vr.similarity_score, top_evidence_score * 0.8), 4) if evidence_grounded else round(vr.similarity_score, 4),
entailment_label=vr.entailment_label if vr.is_supported else ("EVIDENCE_GROUNDED" if evidence_grounded else vr.entailment_label),
best_evidence=vr.best_evidence[:500] if vr.best_evidence else "",
evidence_source=vr.evidence_source,
))
total_claims = result.total_claims if result.total_claims > 0 else 1
boosted_ratio = boosted_supported / total_claims
is_verified = boosted_ratio >= p.firewall_threshold
evidence = []
for ev in result.retrieved_evidence:
evidence.append(EvidenceResult(
content=ev.content[:400],
similarity_score=round(ev.similarity_score, 4),
source=os.path.basename(ev.metadata.get("source", "Unknown")),
rank=ev.rank,
))
# Strip any [Source: ...] tags that leaked into the response
clean_response = re.sub(r'\[Source:\s*[^\]]*\]\s*', '', result.final_response).strip()
# ββ Build prompt refinement suggestion for unverified responses βββ
prompt_suggestion = ""
if not is_verified and total_claims > 0:
unsupported_texts = [c.text for c in claims if not c.is_supported]
if unsupported_texts:
prompt_suggestion = (
f"Try rephrasing your query to be more specific. "
f"For example, try asking about specific topics mentioned in your document. "
f"Unsupported claims: {'; '.join(unsupported_texts[:3])}"
)
clean_response = (
f"{clean_response}\n\n"
f"Verification note: {boosted_supported} of {total_claims} claim(s) were verified. "
f"{total_claims - boosted_supported} claim(s) could not be fully verified."
)
if prompt_suggestion:
clean_response += f"\n\nSuggested refinement: {prompt_suggestion}"
return QueryResponse(
query=req.query,
response=clean_response,
is_verified=is_verified,
support_ratio=round(boosted_ratio, 4),
total_claims=total_claims,
supported_claims=boosted_supported,
regeneration_attempts=result.regeneration_attempts,
claims=claims,
evidence=evidence,
elapsed_seconds=round(elapsed, 3),
)
@app.post("/api/verify", response_model=VerifyResponse)
def verify_claims(req: VerifyRequest):
from core.claim_extractor import Claim
p = get_pipeline()
combined = " ".join(req.claims)
evidence_list = p.retriever.retrieve(combined, top_k=req.top_k)
claim_objs = [Claim(text=t, claim_id=i) for i, t in enumerate(req.claims)]
vr_list = p.verifier.verify_all_claims(claim_objs, evidence_list)
results = []
for vr in vr_list:
results.append(VerifyClaimResult(
text=vr.claim.text,
is_supported=vr.is_supported,
similarity_score=round(vr.similarity_score, 4),
entailment_label=vr.entailment_label,
best_evidence=vr.best_evidence[:500] if vr.best_evidence else "",
))
supported = sum(1 for r in results if r.is_supported)
total = len(results)
return VerifyResponse(
results=results,
supported=supported,
total=total,
ratio=round(supported / total, 4) if total > 0 else 0,
)
@app.post("/api/upload")
async def upload_document(file: UploadFile = File(...)):
"""Upload and ingest a document (TXT, PDF, DOCX, Excel, CSV)."""
ext = os.path.splitext(file.filename)[1].lower()
if ext not in ALLOWED_EXTENSIONS:
raise HTTPException(400, f"Unsupported file type: {ext}. Allowed: {', '.join(ALLOWED_EXTENSIONS)}")
save_path = os.path.join(UPLOAD_DIR, file.filename)
content = await file.read()
with open(save_path, "wb") as f:
f.write(content)
try:
p = get_pipeline()
chunks_added = p.ingest_file(save_path)
uploaded_files.append(file.filename)
# Also load into structured data store for Excel/CSV analytical queries
if ext in (".xlsx", ".xls"):
data_store.load_excel(save_path)
elif ext == ".csv":
data_store.load_csv(save_path)
return {
"filename": file.filename,
"file_type": ext,
"chunks_added": chunks_added,
"total_chunks": p.document_count,
}
except Exception as e:
raise HTTPException(500, f"Failed to process {file.filename}: {str(e)}")
@app.post("/api/clear-uploads")
def clear_uploads():
"""Clear all uploaded documents and reset the pipeline."""
global pipeline, uploaded_files
pipeline = None
uploaded_files = []
data_store.clear()
if os.path.exists(UPLOAD_DIR):
for f in os.listdir(UPLOAD_DIR):
os.remove(os.path.join(UPLOAD_DIR, f))
p = get_pipeline(force_clean=True)
return {"status": "cleared", "total_chunks": p.document_count}
class DeleteRequest(BaseModel):
filename: str
@app.post("/api/delete-file")
def delete_file(req: DeleteRequest):
"""Delete a single uploaded file and rebuild the pipeline."""
global pipeline, uploaded_files
fpath = os.path.join(UPLOAD_DIR, req.filename)
if not os.path.exists(fpath):
raise HTTPException(404, f"File not found: {req.filename}")
# Remove the file
os.remove(fpath)
# Remove from tracked list
uploaded_files = [f for f in uploaded_files if f != req.filename]
# Rebuild pipeline + data store from remaining files
pipeline = None
data_store.clear()
p = get_pipeline(force_clean=True)
return {
"deleted": req.filename,
"remaining_files": uploaded_files,
"total_chunks": p.document_count,
}
# ββ Pre-load pipeline at startup βββββββββββββββββββββββββββββββββββββββββββββ
import threading
def _preload_pipeline():
"""Load ML models in background so first request is fast."""
print("Pre-loading VDHF pipeline (this may take a minute)...")
get_pipeline()
print("Pipeline ready!")
@app.on_event("startup")
def startup_event():
threading.Thread(target=_preload_pipeline, daemon=True).start()
# ββ Serve React build ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
frontend_dist = os.path.join(os.path.dirname(__file__), "frontend", "dist")
if os.path.exists(frontend_dist):
from fastapi.responses import FileResponse
@app.get("/")
def serve_root():
return FileResponse(os.path.join(frontend_dist, "index.html"))
app.mount("/", StaticFiles(directory=frontend_dist, html=True), name="frontend")
|