π Legal Document RAG
Advanced Document Processing with FAISS + BM25 + DeepSeek LLM Enrichment
π Status
π€ Upload
π View Chunks
β¨ Enrichment
π View Chunks
Loading...
β¨ LLM Enrichment
Use DeepSeek LLM to generate summaries, key points, and themes for all chunks.
'''
# ============================================================================
# Confidence Calculation Helpers for Progressive Disclosure UI
# ============================================================================
def calculate_precedential_confidence(boost: float) -> dict:
"""Calculate precedential strength confidence (0-100) with label"""
if boost >= 2.0:
return {'score': 95, 'label': 'Gold Standard', 'description': 'Supreme Court or Restatement authority'}
elif boost >= 1.5:
return {'score': 80, 'label': 'Highly Persuasive', 'description': 'Circuit Court or major treatise'}
elif boost >= 1.2:
return {'score': 65, 'label': 'Persuasive', 'description': 'District Court or state court authority'}
elif boost >= 1.0:
return {'score': 50, 'label': 'Standard', 'description': 'General legal authority'}
else:
return {'score': 25, 'label': 'Declining', 'description': 'May be overruled or superseded'}
def calculate_jurisdiction_confidence(boost: float) -> dict:
"""Calculate jurisdictional relevance confidence"""
if boost >= 1.5:
return {'score': 95, 'label': 'Supreme Court', 'description': 'Binding precedent nationwide'}
elif boost >= 1.3:
return {'score': 80, 'label': 'Federal Circuit', 'description': 'Binding in circuit'}
elif boost >= 1.2:
return {'score': 70, 'label': 'State Supreme', 'description': 'Binding in state'}
elif boost >= 1.1:
return {'score': 60, 'label': 'State Appeals', 'description': 'Persuasive in state'}
else:
return {'score': 50, 'label': 'General', 'description': 'General jurisdiction'}
def calculate_argument_confidence(boost: float) -> dict:
"""Calculate argument structure quality confidence"""
if boost >= 1.4:
return {'score': 90, 'label': 'Strong Structure', 'description': 'Clear analogical or doctrinal framework'}
elif boost >= 1.3:
return {'score': 75, 'label': 'Good Structure', 'description': 'Well-organized legal argument'}
elif boost >= 1.15:
return {'score': 60, 'label': 'Moderate', 'description': 'Standard argument structure'}
else:
return {'score': 45, 'label': 'Basic', 'description': 'Simple presentation'}
def calculate_overall_confidence(result: dict) -> dict:
"""Calculate overall confidence score for a search result"""
boosts = result.get('boosts', {})
# Weight different factors
factors = {
'precedential': (boosts.get('precedential', 1.0), 0.25),
'jurisdiction': (boosts.get('jurisdiction', 1.0), 0.20),
'citation': (boosts.get('citation', 1.0), 0.15),
'argument': (boosts.get('argument', 1.0), 0.15),
'doctrine': (boosts.get('doctrine', 1.0), 0.10),
'treatment': (boosts.get('treatment', 1.0), 0.10),
'temporal': (boosts.get('temporal', 1.0), 0.05)
}
# Calculate weighted score
weighted_sum = 0
for factor, (boost, weight) in factors.items():
# Normalize boost to 0-100 scale
normalized = min(100, max(0, (boost - 0.4) / 1.6 * 100))
weighted_sum += normalized * weight
score = int(weighted_sum)
# Determine label
if score >= 85:
label = 'Very High'
description = 'Highly authoritative and relevant'
elif score >= 70:
label = 'High'
description = 'Strong legal authority'
elif score >= 55:
label = 'Moderate'
description = 'Relevant but verify with additional sources'
elif score >= 40:
label = 'Low'
description = 'Limited authority, use with caution'
else:
label = 'Very Low'
description = 'Weak authority, requires verification'
return {
'score': score,
'label': label,
'description': description,
'factors': {k: round(v[0], 2) for k, v in factors.items()}
}
@app.route('/health', methods=['GET'])
def health():
"""Health check"""
return jsonify({
'status': 'healthy',
'total_documents': len(vector_store.documents),
'has_faiss': HAS_FAISS,
'has_sentence_transformers': HAS_SENTENCE_TRANSFORMERS,
'has_pymupdf': HAS_PYMUPDF
})
@app.route('/status', methods=['GET'])
def status():
"""Get system status"""
# Count unique sources
sources = set(d.get('source', '') for d in vector_store.documents)
# Get LLM service info (via enrichment_service)
llm = enrichment_service.llm
llm_info = {
'chat_llm': 'Grok API (grok-4-fast-reasoning)',
'enrichment_llm': 'DeepSeek (local)' if llm.deepseek_available else 'Grok API',
'deepseek_available': llm.deepseek_available,
'deepseek_model': llm.deepseek_model if llm.deepseek_available else None,
'grok_available': bool(llm.grok_api_key)
}
return jsonify({
'total_documents': len(vector_store.documents),
'total_files': len(sources),
'has_faiss': HAS_FAISS,
'has_sentence_transformers': HAS_SENTENCE_TRANSFORMERS,
'embedding_model': 'SentenceTransformer' if HAS_SENTENCE_TRANSFORMERS else 'hash-based',
'search_type': 'FAISS + BM25 Hybrid' if HAS_FAISS else 'Cosine + BM25 Hybrid',
'llm': llm_info
})
@app.route('/upload', methods=['POST'])
@require_api_key
def upload_file():
"""Upload and process a document"""
if 'file' not in request.files:
return jsonify({'error': 'No file provided'}), 400
file = request.files['file']
if not file.filename:
return jsonify({'error': 'Empty filename'}), 400
# Validate extension
ext = Path(file.filename).suffix.lower()
if ext not in ['.pdf', '.txt']:
return jsonify({'error': 'Only PDF and TXT files are supported'}), 400
# Save file
safe_filename = secure_filename(file.filename)
file_path = config.pdf_dir / safe_filename
file.save(str(file_path))
# Start background processing
job_id = str(uuid.uuid4())[:8]
update_progress(job_id, 'starting', 0, 'File uploaded, starting processing...')
thread = threading.Thread(
target=process_file_background,
args=(job_id, file_path, safe_filename),
daemon=True
)
thread.start()
return jsonify({
'success': True,
'job_id': job_id,
'message': 'File uploaded. Processing started.',
'filename': safe_filename
}), 202
@app.route('/progress/