File size: 14,034 Bytes
d520909 |
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 |
"""
Document Intelligence Bridge for RAG
Bridges the document_intelligence subsystem with the RAG indexer/retriever.
Converts ParseResult to a format compatible with DocumentIndexer.
"""
from typing import List, Optional, Dict, Any
from pathlib import Path
from pydantic import BaseModel
from loguru import logger
from .store import VectorStore, get_vector_store
from .embeddings import EmbeddingAdapter, get_embedding_adapter
from .indexer import IndexingResult, IndexerConfig
# Try to import document_intelligence types
try:
from ..document_intelligence.chunks import (
ParseResult,
DocumentChunk,
BoundingBox,
EvidenceRef,
ChunkType,
)
DOCINT_AVAILABLE = True
except ImportError:
DOCINT_AVAILABLE = False
logger.warning("document_intelligence module not available")
class DocIntIndexer:
"""
Indexes ParseResult from document_intelligence into the vector store.
This bridges the new document_intelligence subsystem with the existing
RAG infrastructure.
"""
def __init__(
self,
config: Optional[IndexerConfig] = None,
vector_store: Optional[VectorStore] = None,
embedding_adapter: Optional[EmbeddingAdapter] = None,
):
self.config = config or IndexerConfig()
self._store = vector_store
self._embedder = embedding_adapter
@property
def store(self) -> VectorStore:
if self._store is None:
self._store = get_vector_store()
return self._store
@property
def embedder(self) -> EmbeddingAdapter:
if self._embedder is None:
self._embedder = get_embedding_adapter()
return self._embedder
def index_parse_result(
self,
parse_result: "ParseResult",
source_path: Optional[str] = None,
) -> IndexingResult:
"""
Index a ParseResult from document_intelligence.
Args:
parse_result: ParseResult from DocumentParser
source_path: Optional override for source path
Returns:
IndexingResult with indexing stats
"""
if not DOCINT_AVAILABLE:
return IndexingResult(
document_id="unknown",
source_path="unknown",
num_chunks_indexed=0,
num_chunks_skipped=0,
success=False,
error="document_intelligence module not available",
)
document_id = parse_result.doc_id
source = source_path or parse_result.filename
try:
chunks_to_index = []
skipped = 0
for chunk in parse_result.chunks:
# Skip empty or short chunks
if self.config.skip_empty_chunks:
if not chunk.text or len(chunk.text.strip()) < self.config.min_chunk_length:
skipped += 1
continue
chunk_data = {
"chunk_id": chunk.chunk_id,
"document_id": document_id,
"source_path": source,
"text": chunk.text,
"sequence_index": chunk.sequence_index,
"confidence": chunk.confidence,
}
if self.config.include_page:
chunk_data["page"] = chunk.page
if self.config.include_chunk_type:
chunk_data["chunk_type"] = chunk.chunk_type.value
if self.config.include_bbox and chunk.bbox:
chunk_data["bbox"] = {
"x_min": chunk.bbox.x_min,
"y_min": chunk.bbox.y_min,
"x_max": chunk.bbox.x_max,
"y_max": chunk.bbox.y_max,
}
chunks_to_index.append(chunk_data)
if not chunks_to_index:
return IndexingResult(
document_id=document_id,
source_path=source,
num_chunks_indexed=0,
num_chunks_skipped=skipped,
success=True,
)
# Generate embeddings in batches
logger.info(f"Generating embeddings for {len(chunks_to_index)} chunks")
texts = [c["text"] for c in chunks_to_index]
embeddings = []
batch_size = self.config.batch_size
for i in range(0, len(texts), batch_size):
batch = texts[i:i + batch_size]
batch_embeddings = self.embedder.embed_batch(batch)
embeddings.extend(batch_embeddings)
# Store in vector database
logger.info(f"Storing {len(chunks_to_index)} chunks in vector store")
self.store.add_chunks(chunks_to_index, embeddings)
logger.info(
f"Indexed document {document_id}: "
f"{len(chunks_to_index)} chunks, {skipped} skipped"
)
return IndexingResult(
document_id=document_id,
source_path=source,
num_chunks_indexed=len(chunks_to_index),
num_chunks_skipped=skipped,
success=True,
)
except Exception as e:
logger.error(f"Failed to index parse result: {e}")
return IndexingResult(
document_id=document_id,
source_path=source,
num_chunks_indexed=0,
num_chunks_skipped=0,
success=False,
error=str(e),
)
def index_document(
self,
path: str,
max_pages: Optional[int] = None,
) -> IndexingResult:
"""
Parse and index a document in one step.
Args:
path: Path to document file
max_pages: Optional limit on pages to process
Returns:
IndexingResult
"""
if not DOCINT_AVAILABLE:
return IndexingResult(
document_id=str(path),
source_path=str(path),
num_chunks_indexed=0,
num_chunks_skipped=0,
success=False,
error="document_intelligence module not available",
)
try:
from ..document_intelligence import DocumentParser, ParserConfig
config = ParserConfig(max_pages=max_pages)
parser = DocumentParser(config=config)
logger.info(f"Parsing document: {path}")
parse_result = parser.parse(path)
return self.index_parse_result(parse_result, source_path=str(path))
except Exception as e:
logger.error(f"Failed to parse and index document: {e}")
return IndexingResult(
document_id=str(path),
source_path=str(path),
num_chunks_indexed=0,
num_chunks_skipped=0,
success=False,
error=str(e),
)
def delete_document(self, document_id: str) -> int:
"""Remove a document from the index."""
return self.store.delete_document(document_id)
def get_stats(self) -> Dict[str, Any]:
"""Get indexing statistics."""
total_chunks = self.store.count()
return {
"total_chunks": total_chunks,
"embedding_model": self.embedder.model_name,
"embedding_dimension": self.embedder.embedding_dimension,
}
class DocIntRetriever:
"""
Retriever with document_intelligence EvidenceRef support.
Wraps DocumentRetriever with conversions to document_intelligence types.
"""
def __init__(
self,
vector_store: Optional[VectorStore] = None,
embedding_adapter: Optional[EmbeddingAdapter] = None,
similarity_threshold: float = 0.5,
):
self._store = vector_store
self._embedder = embedding_adapter
self.similarity_threshold = similarity_threshold
@property
def store(self) -> VectorStore:
if self._store is None:
self._store = get_vector_store()
return self._store
@property
def embedder(self) -> EmbeddingAdapter:
if self._embedder is None:
self._embedder = get_embedding_adapter()
return self._embedder
def retrieve(
self,
query: str,
top_k: int = 5,
document_id: Optional[str] = None,
chunk_types: Optional[List[str]] = None,
page_range: Optional[tuple] = None,
) -> List[Dict[str, Any]]:
"""
Retrieve relevant chunks.
Args:
query: Search query
top_k: Number of results
document_id: Filter by document
chunk_types: Filter by chunk type(s)
page_range: Filter by page range (start, end)
Returns:
List of chunk dicts with metadata
"""
# Build filters
filters = {}
if document_id:
filters["document_id"] = document_id
if chunk_types:
filters["chunk_type"] = chunk_types
if page_range:
filters["page"] = {"min": page_range[0], "max": page_range[1]}
# Embed query
query_embedding = self.embedder.embed_text(query)
# Search
results = self.store.search(
query_embedding=query_embedding,
top_k=top_k,
filters=filters if filters else None,
)
# Convert to dicts
chunks = []
for result in results:
if result.similarity < self.similarity_threshold:
continue
chunk = {
"chunk_id": result.chunk_id,
"document_id": result.document_id,
"text": result.text,
"similarity": result.similarity,
"page": result.page,
"chunk_type": result.chunk_type,
"bbox": result.bbox,
"source_path": result.metadata.get("source_path"),
"confidence": result.metadata.get("confidence"),
}
chunks.append(chunk)
return chunks
def retrieve_with_evidence(
self,
query: str,
top_k: int = 5,
document_id: Optional[str] = None,
chunk_types: Optional[List[str]] = None,
page_range: Optional[tuple] = None,
) -> tuple:
"""
Retrieve chunks with EvidenceRef objects.
Returns:
Tuple of (chunks, evidence_refs)
"""
chunks = self.retrieve(
query, top_k, document_id, chunk_types, page_range
)
evidence_refs = []
if DOCINT_AVAILABLE:
for chunk in chunks:
bbox = None
if chunk.get("bbox"):
bbox_data = chunk["bbox"]
bbox = BoundingBox(
x_min=bbox_data.get("x_min", 0),
y_min=bbox_data.get("y_min", 0),
x_max=bbox_data.get("x_max", 1),
y_max=bbox_data.get("y_max", 1),
normalized=True,
)
else:
bbox = BoundingBox(x_min=0, y_min=0, x_max=1, y_max=1)
evidence = EvidenceRef(
chunk_id=chunk["chunk_id"],
doc_id=chunk["document_id"],
page=chunk.get("page", 1),
bbox=bbox,
source_type=chunk.get("chunk_type", "text"),
snippet=chunk["text"][:200],
confidence=chunk.get("confidence", chunk["similarity"]),
)
evidence_refs.append(evidence)
return chunks, evidence_refs
def build_context(
self,
chunks: List[Dict[str, Any]],
max_length: int = 8000,
) -> str:
"""Build context string from retrieved chunks."""
if not chunks:
return ""
parts = []
for i, chunk in enumerate(chunks, 1):
header = f"[{i}]"
if chunk.get("page"):
header += f" Page {chunk['page']}"
if chunk.get("chunk_type"):
header += f" ({chunk['chunk_type']})"
header += f" [sim={chunk['similarity']:.2f}]"
parts.append(header)
parts.append(chunk["text"])
parts.append("")
context = "\n".join(parts)
if len(context) > max_length:
context = context[:max_length] + "\n...[truncated]"
return context
# Singleton instances
_docint_indexer: Optional[DocIntIndexer] = None
_docint_retriever: Optional[DocIntRetriever] = None
def get_docint_indexer(
config: Optional[IndexerConfig] = None,
vector_store: Optional[VectorStore] = None,
embedding_adapter: Optional[EmbeddingAdapter] = None,
) -> DocIntIndexer:
"""Get or create singleton DocIntIndexer."""
global _docint_indexer
if _docint_indexer is None:
_docint_indexer = DocIntIndexer(
config=config,
vector_store=vector_store,
embedding_adapter=embedding_adapter,
)
return _docint_indexer
def get_docint_retriever(
vector_store: Optional[VectorStore] = None,
embedding_adapter: Optional[EmbeddingAdapter] = None,
similarity_threshold: float = 0.5,
) -> DocIntRetriever:
"""Get or create singleton DocIntRetriever."""
global _docint_retriever
if _docint_retriever is None:
_docint_retriever = DocIntRetriever(
vector_store=vector_store,
embedding_adapter=embedding_adapter,
similarity_threshold=similarity_threshold,
)
return _docint_retriever
def reset_docint_components():
"""Reset singleton instances."""
global _docint_indexer, _docint_retriever
_docint_indexer = None
_docint_retriever = None
|