Spaces:
Sleeping
Sleeping
File size: 21,784 Bytes
a6680e7 ecab17a 292292a 5f78fd3 4f6cdd1 292292a 5f78fd3 4f6cdd1 ecab17a 292292a e3fc89e a6680e7 292292a fa76eb3 ecab17a fa76eb3 292292a a6680e7 ecab17a a6680e7 fa76eb3 ecab17a 292292a ecab17a 292292a fa76eb3 a6680e7 ecab17a fa76eb3 a6680e7 fa76eb3 a6680e7 fa76eb3 292292a fa76eb3 292292a ecab17a 5f78fd3 a6680e7 5f78fd3 a6680e7 5f78fd3 292292a 4f6cdd1 a6680e7 5f78fd3 292292a a6680e7 5f78fd3 292292a a6680e7 292292a a6680e7 5f78fd3 a6680e7 34bfedc 5f78fd3 292292a a6680e7 4f6cdd1 5f78fd3 292292a 4f6cdd1 5f78fd3 292292a a6680e7 5f78fd3 292292a 4f6cdd1 292292a a6680e7 4f6cdd1 5f78fd3 a6680e7 5f78fd3 292292a 5f78fd3 292292a 5f78fd3 a6680e7 5f78fd3 292292a a6680e7 5f78fd3 292292a 5f78fd3 a6680e7 5f78fd3 292292a 5f78fd3 4f6cdd1 a6680e7 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 5f78fd3 4f6cdd1 292292a 4f6cdd1 34bfedc a6680e7 292292a 34bfedc 4f6cdd1 292292a a6680e7 292292a 4f6cdd1 292292a 4f6cdd1 5f78fd3 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 ecab17a 4f6cdd1 a6680e7 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 34bfedc a6680e7 292292a 34bfedc 4f6cdd1 292292a a6680e7 292292a 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 ecab17a 4f6cdd1 292292a 4f6cdd1 a6680e7 292292a a6680e7 292292a 4f6cdd1 5f78fd3 4f6cdd1 292292a a6680e7 292292a 5f78fd3 292292a a6680e7 4f6cdd1 5f78fd3 292292a 5f78fd3 4f6cdd1 292292a 5f78fd3 a6680e7 292292a 5f78fd3 4f6cdd1 292292a 5f78fd3 4f6cdd1 5f78fd3 a6680e7 4f6cdd1 a6680e7 292292a a6680e7 292292a 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 fa76eb3 292292a 4f6cdd1 a6680e7 292292a 4f6cdd1 a6680e7 4f6cdd1 a6680e7 292292a a6680e7 292292a 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 292292a 4f6cdd1 a6680e7 292292a 4f6cdd1 a6680e7 4f6cdd1 a6680e7 292292a a6680e7 292292a a6680e7 292292a 4f6cdd1 292292a 4f6cdd1 292292a 77f58e2 a6680e7 292292a 77f58e2 292292a 77f58e2 a6680e7 77f58e2 292292a 77f58e2 292292a 77f58e2 a6680e7 77f58e2 a6680e7 77f58e2 a6680e7 77f58e2 292292a 77f58e2 292292a 77f58e2 292292a 77f58e2 a6680e7 292292a a6680e7 292292a a6680e7 292292a a6680e7 292292a a6680e7 77f58e2 a6680e7 292292a 77f58e2 292292a 77f58e2 292292a a6680e7 77f58e2 292292a 77f58e2 292292a 77f58e2 292292a 77f58e2 292292a 34bfedc 292292a a6680e7 34bfedc 292292a 34bfedc 77f58e2 292292a 34bfedc 77f58e2 292292a 34bfedc a6680e7 34bfedc 292292a a6680e7 292292a 77f58e2 a6680e7 77f58e2 292292a a6680e7 77f58e2 292292a a6680e7 292292a 77f58e2 292292a 77f58e2 292292a 77f58e2 292292a 77f58e2 a6680e7 292292a 77f58e2 292292a 77f58e2 a6680e7 77f58e2 292292a a6680e7 77f58e2 292292a a6680e7 77f58e2 292292a a6680e7 77f58e2 34bfedc 292292a a6680e7 292292a a6680e7 34bfedc 292292a 34bfedc 292292a a6680e7 292292a a6680e7 292292a a6680e7 292292a a6680e7 292292a a6680e7 292292a a6680e7 292292a a6680e7 292292a a6680e7 |
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 |
"""
Enhanced RAG System - Visual Image Analysis
Sends base64 images directly to GPT-4o for visual analysis (not just OCR)
Then stores results in vector store
"""
from typing import List, Dict
from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage, SystemMessage
import base64
import os
from pathlib import Path
from config import (
OPENAI_API_KEY, OPENAI_MODEL, TEMPERATURE, MAX_TOKENS,
LANGUAGE, CHROMA_DB_PATH
)
class VisualMultimodalRAG:
"""
RAG system that:
1. Sends images as base64 to GPT-4o for visual analysis
2. Gets detailed visual descriptions and insights
3. Stores visual analysis in vector store
4. Enables image-based semantic search
"""
def __init__(self, api_key: str = None, debug: bool = True):
api_key = api_key or OPENAI_API_KEY
self.debug = debug
# Use gpt-4o for vision capabilities
self.llm = ChatOpenAI(
model_name="gpt-4o-mini", # CRITICAL: gpt-4o has vision
api_key=api_key,
temperature=TEMPERATURE,
max_tokens=MAX_TOKENS,
)
self.language = LANGUAGE
self.visual_summaries_log = []
if self.debug:
print("β
VisualMultimodalRAG initialized with gpt-4o (vision model)")
def _debug_print(self, label: str, data: any):
"""Print debug information"""
if self.debug:
print(f"\nπ DEBUG [{label}]:")
if isinstance(data, (list, dict)):
print(f" Type: {type(data).__name__}")
print(f" Content: {str(data)[:300]}...")
else:
print(f" {data}")
def _image_to_base64(self, image_path: str) -> str:
"""Convert image file to base64 string"""
try:
with open(image_path, 'rb') as image_file:
image_data = base64.b64encode(image_file.read()).decode('utf-8')
return image_data
except Exception as e:
print(f"Error converting image to base64: {e}")
return None
def analyze_image_visually(self, image_path: str, image_idx: int) -> str:
"""
Send actual image (base64) to gpt-4o for visual analysis
Returns detailed visual analysis/description
gpt-4o can see:
- Charts, graphs, diagrams
- Tables and structured data
- Photos and drawings
- Handwritten text
- Screenshots
- Any visual content
"""
if not os.path.exists(image_path):
return f"[Image {image_idx}: File not found - {image_path}]"
try:
# Convert image to base64
image_base64 = self._image_to_base64(image_path)
if not image_base64:
return f"[Image {image_idx}: Could not convert to base64]"
# Determine image type
file_ext = Path(image_path).suffix.lower()
media_type_map = {
'.jpg': 'image/jpeg',
'.jpeg': 'image/jpeg',
'.png': 'image/png',
'.gif': 'image/gif',
'.webp': 'image/webp'
}
media_type = media_type_map.get(file_ext, 'image/png')
print(f"π Analyzing image {image_idx} visually (as {media_type})...")
# Create message with image
message = HumanMessage(
content=[
{
"type": "image_url",
"image_url": {
"url": f"data:{media_type};base64,{image_base64}",
},
},
{
"type": "text",
"text": f"""Analyze this image in detail in {self.language}.
Provide a comprehensive visual analysis including:
1. **What you see** - Main objects, elements, structure
2. **Data/Content** - Any numbers, text, charts, graphs
3. **Purpose** - What this image is showing or representing
4. **Key insights** - Important patterns, trends, or information
5. **Connections** - How this relates to document content
Be specific and detailed. Focus on visual information that cannot be extracted from text alone.
Analysis:"""
}
],
)
# Call gpt-4o with vision
response = self.llm.invoke([message])
analysis = response.content.strip()
if self.debug:
self._debug_print(f"Image {image_idx} Visual Analysis", analysis)
print(f"β
Image {image_idx} analyzed successfully")
return analysis
except Exception as e:
error_msg = f"[Image {image_idx}: Vision analysis failed - {str(e)}]"
print(f"β Error analyzing image {image_idx}: {e}")
return error_msg
def analyze_images_visually(self, images: List[Dict]) -> List[Dict]:
"""
Analyze each image visually using gpt-4o vision
Returns list of {image_index, visual_analysis, type}
"""
visual_analyses = []
for idx, image in enumerate(images):
image_path = image.get('path', '')
if not image_path:
print(f"β οΈ Image {idx}: No path provided")
continue
# Analyze image visually (not just OCR)
visual_analysis = self.analyze_image_visually(image_path, idx)
visual_analyses.append({
'type': 'image_visual',
'image_index': idx,
'image_path': image_path,
'visual_analysis': visual_analysis,
'ocr_text': image.get('ocr_text', '') # Keep OCR as backup
})
return visual_analyses
def summarize_text_chunks(self, text: str, chunk_size: int = 1500) -> List[Dict]:
"""
Chunk text and summarize each chunk individually
"""
chunks = []
text_chunks = self._chunk_text(text, chunk_size=chunk_size, overlap=300)
self._debug_print("Text Chunking", f"Created {len(text_chunks)} chunks")
for idx, chunk in enumerate(text_chunks):
if len(chunk.strip()) < 50:
continue
try:
prompt = f"""Summarize this text chunk in {self.language}.
Keep it concise. Extract key points, facts, and main ideas.
Text Chunk:
{chunk}
Summary (2-3 sentences maximum):"""
message = HumanMessage(content=prompt)
response = self.llm.invoke([message])
summary = response.content.strip()
chunks.append({
'type': 'text_chunk',
'chunk_index': len(chunks),
'original_text': chunk[:500],
'summary': summary,
'chunk_length': len(chunk)
})
if self.debug:
self._debug_print(f"Text Chunk {len(chunks)-1} Summary", summary)
except Exception as e:
print(f"Error summarizing text chunk: {e}")
return chunks
def summarize_tables(self, tables: List[Dict]) -> List[Dict]:
"""
Summarize each table individually
"""
summaries = []
for idx, table in enumerate(tables):
table_content = table.get('content', '')
if not table_content or len(table_content.strip()) < 10:
continue
try:
prompt = f"""Analyze and summarize this table/structured data in {self.language}.
Extract key insights, row/column meanings, and important figures.
Table Content:
{table_content}
Summary (2-3 sentences maximum):"""
message = HumanMessage(content=prompt)
response = self.llm.invoke([message])
summary = response.content.strip()
summaries.append({
'type': 'table',
'table_index': idx,
'original_content': table_content[:500],
'summary': summary,
'table_length': len(table_content)
})
if self.debug:
self._debug_print(f"Table {idx} Summary", summary)
except Exception as e:
print(f"Error summarizing table {idx}: {e}")
return summaries
def process_and_store_document(
self,
text: str,
images: List[Dict],
tables: List[Dict],
vector_store,
doc_id: str
) -> Dict:
"""
Main function: Analyze all components visually and store in vector store
Images are analyzed using gpt-4o vision (not just OCR)
"""
print(f"\n{'='*70}")
print(f"PROCESSING WITH VISUAL IMAGE ANALYSIS: {doc_id}")
print(f"{'='*70}")
results = {
'doc_id': doc_id,
'image_visual_analyses': [],
'text_summaries': [],
'table_summaries': [],
'total_stored': 0
}
# 1. Analyze images VISUALLY using gpt-4o
print(f"\nπΌοΈ VISUAL IMAGE ANALYSIS (gpt-4o vision) ({len(images)} total)")
print(f"{'β'*70}")
image_analyses = self.analyze_images_visually(images)
results['image_visual_analyses'] = image_analyses
# Store each image analysis in vector store
image_docs = {
'text': ' | '.join([
f"Image {a['image_index']}: {a['visual_analysis']}"
for a in image_analyses
]),
'images': [],
'tables': []
}
for analysis in image_analyses:
print(f" β
Image {analysis['image_index']} (visual analysis)")
print(f" Path: {analysis['image_path']}")
print(f" Analysis: {analysis['visual_analysis'][:100]}...")
if image_analyses:
try:
vector_store.add_documents(
image_docs,
f"{doc_id}_images_visual"
)
results['total_stored'] += len(image_analyses)
print(f"β
Stored {len(image_analyses)} image visual analyses")
except Exception as e:
print(f"β Error storing image analyses: {e}")
# 2. Summarize and store text chunks
print(f"\nπ TEXT CHUNK SUMMARIZATION")
print(f"{'β'*70}")
text_summaries = self.summarize_text_chunks(text)
results['text_summaries'] = text_summaries
text_docs = {
'text': ' | '.join([f"Chunk {s['chunk_index']}: {s['summary']}"
for s in text_summaries]),
'images': [],
'tables': []
}
for summary in text_summaries:
print(f" β
Chunk {summary['chunk_index']}: {summary['summary'][:50]}...")
if text_summaries:
try:
vector_store.add_documents(
text_docs,
f"{doc_id}_text_chunks"
)
results['total_stored'] += len(text_summaries)
print(f"β
Stored {len(text_summaries)} text chunk summaries")
except Exception as e:
print(f"β Error storing text summaries: {e}")
# 3. Summarize and store tables
print(f"\nπ TABLE SUMMARIZATION ({len(tables)} total)")
print(f"{'β'*70}")
table_summaries = self.summarize_tables(tables)
results['table_summaries'] = table_summaries
table_docs = {
'text': ' | '.join([f"Table {s['table_index']}: {s['summary']}"
for s in table_summaries]),
'images': [],
'tables': []
}
for summary in table_summaries:
print(f" β
Table {summary['table_index']}: {summary['summary'][:50]}...")
if table_summaries:
try:
vector_store.add_documents(
table_docs,
f"{doc_id}_tables"
)
results['total_stored'] += len(table_summaries)
print(f"β
Stored {len(table_summaries)} table summaries")
except Exception as e:
print(f"β Error storing table summaries: {e}")
# 4. Summary statistics
print(f"\n{'='*70}")
print(f"π STORAGE SUMMARY")
print(f"{'='*70}")
print(f" Images analyzed visually & stored: {len(image_analyses)}")
print(f" Text chunks summarized & stored: {len(text_summaries)}")
print(f" Tables summarized & stored: {len(table_summaries)}")
print(f" Total items stored in vector: {results['total_stored']}")
print(f"{'='*70}")
self.visual_summaries_log.append(results)
return results
def _chunk_text(self, text: str, chunk_size: int = 1500, overlap: int = 300) -> List[str]:
"""Split text into overlapping chunks"""
chunks = []
start = 0
while start < len(text):
end = start + chunk_size
chunks.append(text[start:end])
start = end - overlap
return chunks
def get_visual_summaries_log(self) -> List[Dict]:
"""Get all visual analysis logs"""
return self.visual_summaries_log
class AnsweringRAG:
"""
RAG system that:
1. Searches vector store for relevant content
2. ANALYZES search results
3. Generates intelligent answers based on context
"""
def __init__(self, api_key: str = None, debug: bool = True):
api_key = api_key or OPENAI_API_KEY
self.debug = debug
self.llm = ChatOpenAI(
model_name="gpt-4o-mini", # Use gpt-4o for better understanding
api_key=api_key,
temperature=TEMPERATURE,
max_tokens=MAX_TOKENS,
)
self.language = LANGUAGE
self.answer_log = []
if self.debug:
print("β
AnsweringRAG initialized with answer generation")
def _debug_print(self, label: str, data: any):
"""Print debug information"""
if self.debug:
print(f"\nπ DEBUG [{label}]:")
if isinstance(data, (list, dict)):
print(f" Type: {type(data).__name__}")
print(f" Content: {str(data)[:300]}...")
else:
print(f" {data}")
def analyze_and_answer(
self,
question: str,
search_results: List[Dict]
) -> Dict:
"""
Analyze search results and generate intelligent answer
Returns:
{
'question': user question,
'answer': detailed answer,
'sources_used': number of sources,
'confidence': low/medium/high,
'search_results': original search results
}
"""
print(f"\n{'='*70}")
print(f"ANALYZING QUESTION & GENERATING ANSWER")
print(f"{'='*70}")
print(f"\nβ Question: {question}")
print(f"π Search Results Found: {len(search_results)}")
# Check if we have search results
if not search_results:
print(f"β οΈ No search results found!")
answer = f"""I could not find relevant information in the document to answer your question: "{question}"
Try:
- Using different keywords
- Breaking the question into smaller parts
- Asking about other topics in the document"""
result = {
'question': question,
'answer': answer,
'sources_used': 0,
'confidence': 'low',
'search_results': []
}
self.answer_log.append(result)
return result
# Build context from search results
context_parts = []
for idx, result in enumerate(search_results, 1):
content = result.get('content', '')
metadata = result.get('metadata', {})
content_type = result.get('type', 'unknown')
distance = result.get('distance', 0)
relevance = 1 - distance if distance else 0
context_parts.append(f"""
[Source {idx} - {content_type.upper()} (relevance: {relevance:.1%})]
{content}""")
full_context = "\n".join(context_parts)
self._debug_print("Context Prepared", f"{len(context_parts)} sources, {len(full_context)} chars")
# Build prompt to analyze results and answer question
analysis_prompt = f"""You are a helpful assistant analyzing document content to answer user questions.
USER QUESTION:
"{question}"
RELEVANT CONTENT FROM DOCUMENT:
{full_context}
INSTRUCTIONS:
1. Analyze the provided content carefully
2. Extract information relevant to the question
3. Synthesize a clear, comprehensive answer in {self.language}
4. If the content doesn't fully answer the question, explain what information is available
5. Be specific and cite the content when relevant
6. Structure your answer clearly with key points
ANSWER:"""
print(f"\nπ Analyzing search results...")
print(f" Context size: {len(full_context)} characters")
print(f" Sources: {len(search_results)}")
try:
# Call LLM to analyze and answer
message = HumanMessage(content=analysis_prompt)
response = self.llm.invoke([message])
answer = response.content.strip()
# Determine confidence level
confidence = self._estimate_confidence(len(search_results), answer)
print(f"β
Answer generated successfully")
print(f" Confidence: {confidence}")
print(f" Answer length: {len(answer)} characters")
result = {
'question': question,
'answer': answer,
'sources_used': len(search_results),
'confidence': confidence,
'search_results': search_results
}
self.answer_log.append(result)
return result
except Exception as e:
print(f"β Error generating answer: {e}")
answer = f"I encountered an error while analyzing the search results. Please try again."
result = {
'question': question,
'answer': answer,
'sources_used': len(search_results),
'confidence': 'low',
'error': str(e),
'search_results': search_results
}
self.answer_log.append(result)
return result
def _estimate_confidence(self, sources_count: int, answer: str) -> str:
"""Estimate confidence level of answer"""
answer_length = len(answer)
# High confidence: multiple sources, substantial answer
if sources_count >= 3 and answer_length > 500:
return "high"
# Medium confidence: some sources, decent answer
elif sources_count >= 2 and answer_length > 200:
return "medium"
# Low confidence: few sources or short answer
else:
return "low"
def get_answer_with_sources(
self,
question: str,
search_results: List[Dict]
) -> Dict:
"""
Get answer AND properly formatted sources
Returns both answer and formatted source citations
"""
result = self.analyze_and_answer(question, search_results)
# Format sources for display
formatted_sources = []
for idx, source in enumerate(result['search_results'], 1):
formatted_sources.append({
'index': idx,
'type': source.get('type', 'unknown'),
'content': source.get('content', ''),
'relevance': 1 - source.get('distance', 0) if source.get('distance') else 0
})
result['formatted_sources'] = formatted_sources
return result
def get_answer_log(self) -> List[Dict]:
"""Get all answer generation logs"""
return self.answer_log
def print_answer_with_sources(self, result: Dict, max_source_length: int = 300):
"""Pretty print answer with sources"""
print(f"\n{'='*70}")
print(f"ANSWER TO: {result['question']}")
print(f"{'='*70}")
print(f"\nπ ANSWER (Confidence: {result['confidence'].upper()}):")
print(f"{'-'*70}")
print(result['answer'])
print(f"{'-'*70}")
if result.get('formatted_sources'):
print(f"\nπ SOURCES USED ({len(result['formatted_sources'])} total):")
for source in result['formatted_sources']:
print(f"\n[Source {source['index']} - {source['type'].upper()} ({source['relevance']:.0%} relevant)]")
print(f"{source['content'][:max_source_length]}...")
print(f"\n{'='*70}") |