Spaces:
Sleeping
Sleeping
File size: 25,186 Bytes
a66d4bd | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 | """
RAG Agent - Advanced Retrieval-Augmented Generation Agent
This module implements a RAG Agent that:
- Accepts files uploaded from the frontend via FastAPI
- Processes uploaded files dynamically (PDF, TXT, etc.)
- Creates vector embeddings from uploaded content using Weaviate
- Uses Query Decomposition for focused retrieval
- Uses Reciprocal Rank Fusion (RRF) for intelligent result merging
- Returns responses based on the uploaded file content
Requires Weaviate running on localhost:8081
"""
import logging
import os
import json
import tempfile
import shutil
import re
from typing import Optional, List, Any, Dict
from collections import defaultdict
from pathlib import Path
from dotenv import load_dotenv, find_dotenv
import weaviate
from langchain_community.document_loaders import PyPDFLoader, TextLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_weaviate import WeaviateVectorStore
from langchain_huggingface.embeddings import HuggingFaceEmbeddings
from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.documents import Document
logger = logging.getLogger(__name__)
# Load environment variables
_ = load_dotenv(find_dotenv())
class AdvancedRAGSystem:
"""
Production-ready RAG system with hybrid retrieval + RRF.
Features:
- Hybrid Retrieval: Original query + decomposed sub-queries
- Reciprocal Rank Fusion (RRF): Intelligently merge results
- Keyword Boosting: Prioritize documents with relevant terms
- Cost-efficient: Only 2 LLM calls (decomposition + answer)
- Fully scalable with configurable parameters
"""
def __init__(
self,
vector_store,
llm,
retriever_k: int = 10,
num_sub_queries: int = 2,
rrf_k: int = 60,
keyword_boost: float = 0.25,
top_docs: int = 5
):
"""
Initialize the Advanced RAG System.
Args:
vector_store: Weaviate/Pinecone/etc vector store
llm: Language model for decomposition and answer generation
retriever_k: Number of documents to retrieve per query
num_sub_queries: Number of sub-queries to generate (lower = cheaper)
rrf_k: RRF constant (higher = flatter ranking)
keyword_boost: Boost factor per keyword match
top_docs: Number of top documents for final context
"""
self.vector_store = vector_store
self.llm = llm
self.retriever_k = retriever_k
self.num_sub_queries = num_sub_queries
self.rrf_k = rrf_k
self.keyword_boost = keyword_boost
self.top_docs = top_docs
self.retriever = vector_store.as_retriever(
search_type="similarity",
search_kwargs={"k": retriever_k}
)
self._build_chains()
logger.info(f"AdvancedRAGSystem initialized with k={retriever_k}, sub_queries={num_sub_queries}")
def _build_chains(self):
"""Build the internal LangChain pipelines."""
# Query decomposition prompt
decomposition_template = f"""Rewrite this question into {self.num_sub_queries} specific search queries.
RULES:
1. Include technical keywords that would appear in documentation
2. Focus on syntax, commands, and implementation details
3. Keep the core topic but make it more specific
Question: {{question}}
Write {self.num_sub_queries} search queries (one per line):"""
self.decomposition_prompt = ChatPromptTemplate.from_template(decomposition_template)
# Build decomposer chain
self.query_decomposer = (
self.decomposition_prompt
| self.llm
| StrOutputParser()
| (lambda x: [q.strip() for q in x.strip().split("\n") if q.strip() and len(q.strip()) > 5][:self.num_sub_queries])
)
# RAG answer prompt
self.rag_prompt = ChatPromptTemplate.from_template("""Answer the question using ONLY the provided context.
Context:
{context}
Question: {question}
Instructions:
- Use only information from the context
- If the answer isn't in the context, say "I don't have enough information"
- Be specific and cite relevant details
- Format your answer clearly""")
def _extract_keywords(self, question: str) -> List[str]:
"""Extract keywords from question for boosting."""
stop_words = {'what', 'how', 'why', 'when', 'where', 'is', 'are', 'the',
'a', 'an', 'to', 'in', 'for', 'of', 'and', 'or', 'can', 'do',
'explain', 'describe', 'tell', 'me', 'about'}
words = question.lower().replace('?', '').replace('.', '').split()
keywords = [w for w in words if w not in stop_words and len(w) > 2]
return keywords
def _reciprocal_rank_fusion(self, results: List[List], keywords: List[str] = None) -> List:
"""Apply RRF to merge multiple ranked document lists with keyword boosting."""
fused_scores = defaultdict(float)
doc_map = {}
for doc_list in results:
for rank, doc in enumerate(doc_list):
doc_key = (doc.page_content, json.dumps(doc.metadata, sort_keys=True, default=str))
# Base RRF score: 1 / (k + rank + 1)
score = 1 / (self.rrf_k + rank + 1)
# Apply keyword boost
if keywords:
content_lower = doc.page_content.lower()
matches = sum(1 for kw in keywords if kw in content_lower)
score *= (1 + self.keyword_boost * matches)
fused_scores[doc_key] += score
if doc_key not in doc_map:
doc_map[doc_key] = doc
# Sort by fused score (descending)
reranked = sorted(
[(doc_map[k], s) for k, s in fused_scores.items()],
key=lambda x: x[1],
reverse=True
)
return [doc for doc, _ in reranked]
def _format_context(self, docs: List) -> str:
"""Format documents into context string."""
return "\n\n".join(
f"[Doc {i+1}] {doc.page_content}"
for i, doc in enumerate(docs[:self.top_docs])
)
def retrieve(self, question: str) -> List:
"""
Hybrid retrieval: original query + decomposed queries + RRF.
Args:
question: User's question
Returns:
List of relevant documents ranked by RRF score
"""
keywords = self._extract_keywords(question)
all_results = []
# 1. ALWAYS include original query results
original_docs = self.retriever.invoke(question)
all_results.append(original_docs)
# 2. Add decomposed sub-query results
try:
sub_queries = self.query_decomposer.invoke({"question": question})
for sq in sub_queries:
docs = self.retriever.invoke(sq)
all_results.append(docs)
except Exception as e:
logger.warning(f"Sub-query decomposition skipped: {str(e)[:50]}")
# 3. Apply RRF with keyword boosting
ranked_docs = self._reciprocal_rank_fusion(all_results, keywords)
return ranked_docs
def query(self, question: str) -> str:
"""
Full RAG pipeline: retrieve + generate answer.
Args:
question: User's question
Returns:
Generated answer based on retrieved context
"""
docs = self.retrieve(question)
context = self._format_context(docs)
chain = self.rag_prompt | self.llm | StrOutputParser()
return chain.invoke({"context": context, "question": question})
class RAGAgent:
"""
RAG Agent - Handles document-based question answering with files from frontend.
This agent:
- Receives files uploaded from the frontend via FastAPI
- Processes uploaded files (PDF, TXT, etc.)
- Creates vector embeddings using Weaviate
- Answers questions based on the uploaded file content
"""
def __init__(
self,
weaviate_port: int = 8081,
index_name: str = "UploadedDocuments",
retriever_k: int = 10,
num_sub_queries: int = 2,
chunk_size: int = 1000,
chunk_overlap: int = 200,
):
"""
Initialize the RAG Agent.
Args:
weaviate_port: Port where Weaviate is running
index_name: Name for the Weaviate index
retriever_k: Documents to retrieve per query
num_sub_queries: Sub-queries to generate
chunk_size: Size of text chunks
chunk_overlap: Overlap between chunks
"""
self.weaviate_port = weaviate_port
self.index_name = index_name
self.retriever_k = retriever_k
self.num_sub_queries = num_sub_queries
self.chunk_size = chunk_size
self.chunk_overlap = chunk_overlap
# Will be set when processing a file
self.weaviate_client = None
self.llm = None
self.embeddings = None
# Per-conversation state ("session" here means a chat/conversation id)
# session_id -> {"vector_store": ..., "rag_system": ..., "current_file_name": str, "index_name": str}
self._sessions: Dict[str, Dict[str, Any]] = {}
# Temp directory for uploaded files
self.temp_dir = tempfile.mkdtemp(prefix="rag_uploads_")
# Initialize embeddings and LLM
self._init_embeddings()
self._init_llm()
logger.info("RAG Agent initialized - ready to receive files from frontend")
def _normalize_session_id(self, session_id: Optional[str]) -> str:
"""Normalize a conversation/session id into a safe, stable identifier."""
if not session_id:
return "default"
session_id = str(session_id).strip()
if not session_id:
return "default"
# Allow only safe characters; cap length to avoid huge class names
session_id = re.sub(r"[^a-zA-Z0-9_-]", "_", session_id)[:64]
return session_id or "default"
def _index_name_for_session(self, session_id: str) -> str:
"""Build a Weaviate index/class name for a session."""
session_id = self._normalize_session_id(session_id)
# Keep the base index name stable and ensure it starts with a letter (Weaviate class naming rules)
base = re.sub(r"[^a-zA-Z0-9_]", "_", str(self.index_name)) or "UploadedDocuments"
if not base[0].isalpha():
base = f"C_{base}"
return f"{base}_{session_id}"
def _delete_index_best_effort(self, index_name: str) -> None:
"""Delete a Weaviate collection/index if it exists (best-effort)."""
if self.weaviate_client is None:
return
try:
# Weaviate client v4
self.weaviate_client.collections.delete(index_name)
logger.info(f"Deleted Weaviate index: {index_name}")
except Exception:
# Ignore if it doesn't exist or deletion isn't supported
pass
def _get_session(self, session_id: Optional[str]) -> Dict[str, Any]:
sid = self._normalize_session_id(session_id)
return self._sessions.get(sid, {})
def _init_embeddings(self):
"""Initialize embeddings model."""
try:
logger.info("Loading embeddings model...")
self.embeddings = HuggingFaceEmbeddings(
model_name="sentence-transformers/all-mpnet-base-v2"
)
logger.info("β
Embeddings model loaded")
except Exception as e:
logger.error(f"Failed to load embeddings: {e}")
raise
def _init_llm(self):
"""Initialize LLM."""
try:
logger.info("Initializing LLM for RAG...")
openrouter_api_key = os.getenv("OPENROUTER_API_KEY", "").strip().strip('"').strip("'")
if not openrouter_api_key or openrouter_api_key.startswith("your-"):
raise RuntimeError("Missing or invalid OPENROUTER_API_KEY environment variable")
self.llm = ChatOpenAI(
model="xiaomi/mimo-v2-flash:free",
temperature=0,
openai_api_key=openrouter_api_key,
openai_api_base="https://openrouter.ai/api/v1",
)
logger.info("β
LLM initialized for RAG")
except Exception as e:
logger.error(f"Failed to initialize LLM: {e}")
raise
def _connect_weaviate(self):
"""Connect to Weaviate if not already connected."""
if self.weaviate_client is None:
logger.info(f"Connecting to Weaviate on port {self.weaviate_port}...")
self.weaviate_client = weaviate.connect_to_local(host= "192.168.1.5",port=self.weaviate_port)
if not self.weaviate_client.is_ready():
raise RuntimeError(f"Weaviate is not ready at localhost:{self.weaviate_port}")
logger.info("β
Weaviate connected")
def _load_file(self, file_path: str) -> List[Document]:
"""Load a file and return documents."""
file_ext = Path(file_path).suffix.lower()
if file_ext == ".pdf":
loader = PyPDFLoader(file_path)
elif file_ext in [".txt", ".md", ".py", ".js", ".json", ".csv"]:
loader = TextLoader(file_path, encoding="utf-8")
else:
# Try as text file
loader = TextLoader(file_path, encoding="utf-8")
return loader.load()
def process_file_from_bytes(self, file_content: bytes, filename: str, session_id: Optional[str] = None) -> Dict[str, Any]:
"""
Process a file uploaded from the frontend (synchronous).
Args:
file_content: Raw bytes of the uploaded file
filename: Original filename
Returns:
Dict with status and info about the processed file
"""
try:
session_id = self._normalize_session_id(session_id)
logger.info(f"Processing uploaded file: {filename}")
# Connect to Weaviate
self._connect_weaviate()
# Save file temporarily (avoid trusting user filename for paths)
suffix = Path(filename).suffix if filename else ""
with tempfile.NamedTemporaryFile(delete=False, dir=self.temp_dir, suffix=suffix, prefix="upload_") as tmp:
tmp.write(file_content)
file_path = tmp.name
logger.info(f"File saved to: {file_path}")
# Load documents from file
documents = self._load_file(file_path)
logger.info(f"β
Loaded {len(documents)} pages/sections from {filename}")
# Split into chunks
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=self.chunk_size,
chunk_overlap=self.chunk_overlap
)
docs = text_splitter.split_documents(documents)
logger.info(f"β
Split into {len(docs)} chunks")
# Use a per-session index so multiple conversations don't mix documents.
session_index_name = self._index_name_for_session(session_id)
# Replace any prior session index (ChatGPT-like behavior: latest upload becomes active)
self._delete_index_best_effort(session_index_name)
# Create vector store with Weaviate
logger.info("Creating vector embeddings with Weaviate...")
vector_store = WeaviateVectorStore.from_documents(
documents=docs,
embedding=self.embeddings,
client=self.weaviate_client,
index_name=session_index_name,
text_key="text",
)
logger.info("β
Vector store created with Weaviate")
# Create RAG system
rag_system = AdvancedRAGSystem(
vector_store=vector_store,
llm=self.llm,
retriever_k=self.retriever_k,
num_sub_queries=self.num_sub_queries,
)
# Persist per-session state
self._sessions[session_id] = {
"vector_store": vector_store,
"rag_system": rag_system,
"current_file_name": filename,
"index_name": session_index_name,
}
logger.info(f"β
RAG system ready for session={session_id}, file={filename}")
# Clean up temp file
try:
os.remove(file_path)
except:
pass
return {
"success": True,
"filename": filename,
"session_id": session_id,
"pages": len(documents),
"chunks": len(docs),
"message": f"Successfully processed {filename}. Ready to answer questions."
}
except Exception as e:
logger.error(f"Error processing file {filename}: {e}", exc_info=True)
return {
"success": False,
"filename": filename,
"session_id": session_id,
"error": str(e),
"message": f"Failed to process {filename}: {str(e)}"
}
def initialize(self) -> bool:
"""Initialize RAG Agent - connect to Weaviate."""
try:
self._connect_weaviate()
logger.info("RAG Agent ready (using Weaviate)")
return True
except Exception as e:
logger.error(f"Failed to initialize RAG Agent: {e}")
return False
def retrieve_context(self, question: str, session_id: Optional[str] = None) -> str:
"""
Retrieve relevant context from the uploaded file for a question.
Args:
question: User's question
Returns:
Retrieved context as a string
"""
session_id = self._normalize_session_id(session_id)
rag_system = self._sessions.get(session_id, {}).get("rag_system")
if not rag_system:
return ""
try:
docs = rag_system.retrieve(question)
context = rag_system._format_context(docs)
logger.info(f"Retrieved {len(docs)} relevant chunks for question")
return context
except Exception as e:
logger.error(f"Error retrieving context: {e}")
return ""
def answer_question(self, question: str, session_id: Optional[str] = None) -> str:
"""
Answer a question based on the uploaded file.
Args:
question: User's question about the uploaded file
Returns:
Generated answer based on the file content
"""
session_id = self._normalize_session_id(session_id)
rag_system = self._sessions.get(session_id, {}).get("rag_system")
if not rag_system:
return "No file has been uploaded yet. Please upload a file first before asking questions."
try:
logger.info(f"Processing RAG query: {question[:50]}...")
answer = rag_system.query(question)
logger.info("β
RAG query processed successfully")
return answer
except Exception as e:
logger.error(f"Error processing RAG query: {e}", exc_info=True)
return f"Error processing query: {str(e)}"
def has_file_loaded(self, session_id: Optional[str] = None) -> bool:
"""Check if a file has been processed and is ready for queries (per session)."""
session_id = self._normalize_session_id(session_id)
return bool(self._sessions.get(session_id, {}).get("rag_system"))
def get_current_file(self, session_id: Optional[str] = None) -> Optional[str]:
"""Get the name of the currently loaded file (per session)."""
session_id = self._normalize_session_id(session_id)
return self._sessions.get(session_id, {}).get("current_file_name")
def clear(self, session_id: Optional[str] = None):
"""Clear the current file and vector store for a session."""
session_id = self._normalize_session_id(session_id)
session = self._sessions.pop(session_id, None)
if session and session.get("index_name"):
self._delete_index_best_effort(session["index_name"])
logger.info(f"RAG Agent cleared for session={session_id} - ready for new file")
def close(self):
"""Close connections and cleanup."""
try:
# Close Weaviate connection
if self.weaviate_client is not None:
self.weaviate_client.close()
self.weaviate_client = None
logger.info("β
Weaviate connection closed")
# Clean up temp directory
if os.path.exists(self.temp_dir):
shutil.rmtree(self.temp_dir, ignore_errors=True)
self._sessions.clear()
logger.info("β
RAG Agent cleanup complete")
except Exception as e:
logger.warning(f"Error during cleanup: {e}")
# ============================================================================
# GLOBAL RAG AGENT INSTANCE
# ============================================================================
_rag_agent: Optional[RAGAgent] = None
def get_rag_agent() -> RAGAgent:
"""Get or create the global RAG Agent instance."""
global _rag_agent
if _rag_agent is None:
_rag_agent = RAGAgent()
return _rag_agent
def process_uploaded_file(file_content: bytes, filename: str, session_id: Optional[str] = None) -> Dict[str, Any]:
"""
Process a file uploaded from the frontend.
This function is called by FastAPI when a file is uploaded.
Args:
file_content: Raw bytes of the uploaded file
filename: Original filename
Returns:
Dict with status and info about the processed file
"""
agent = get_rag_agent()
return agent.process_file_from_bytes(file_content, filename, session_id=session_id)
def retrieve_context_for_query(question: str, session_id: Optional[str] = None) -> str:
"""
Retrieve relevant context from uploaded file for a query.
Args:
question: User's question
Returns:
Retrieved context string
"""
agent = get_rag_agent()
return agent.retrieve_context(question, session_id=session_id)
async def answer_rag_question(question: str, session_id: Optional[str] = None) -> str:
"""
Answer a question using the RAG Agent.
Args:
question: User's question
Returns:
RAG-generated answer
"""
agent = get_rag_agent()
return agent.answer_question(question, session_id=session_id)
def has_file_loaded(session_id: Optional[str] = None) -> bool:
"""Check if a file has been loaded into the RAG agent (per session)."""
agent = get_rag_agent()
return agent.has_file_loaded(session_id=session_id)
def cleanup_rag_agent():
"""Cleanup RAG Agent resources."""
global _rag_agent
if _rag_agent is not None:
_rag_agent.close()
_rag_agent = None
logger.info("RAG Agent cleaned up")
# ============================================================================
# FOR TESTING
# ============================================================================
if __name__ == "__main__":
import asyncio
logging.basicConfig(level=logging.INFO)
async def test_rag_agent():
"""Test the RAG Agent with a sample in-memory file."""
print("=" * 80)
print("RAG AGENT TEST")
print("=" * 80)
session_id = "local_test"
sample_content = b"""
Python is a high-level programming language.
It was created by Guido van Rossum in 1991.
Python is known for its simple syntax and readability.
It supports multiple programming paradigms including procedural, object-oriented, and functional programming.
Python has a large standard library and active community.
"""
result = process_uploaded_file(sample_content, "sample.txt", session_id=session_id)
print(f"\nFile processing result: {result}")
if result.get("success"):
question = "Who created Python?"
answer = await answer_rag_question(question, session_id=session_id)
print(f"\nQ: {question}")
print(f"A: {answer}")
cleanup_rag_agent()
asyncio.run(test_rag_agent())
|