Core Features
Qdrant • FalkorDB • LangChain
PostgreSQL • Redis • JWT Auth
LangGraph • GPT-4o-mini
Architecture
Vector Path (Fast)
Direct similarity search for factual queries. ~200-500ms response time.
Use for: Single-concept queries, keyword search, document retrieval
Graph-Lite (Complex)
Entity extraction + graph traversal for relationships. ~500ms-2s response time.
Use for: Multi-hop reasoning, comparisons, trend analysis
Smart Routing
Automatic query classification based on keyword detection and intent analysis.
Use for: Seamless switching between vector and graph paths
How It Works
User submits natural language query through chat interface
Keyword detection classifies as FACTUAL or COMPLEX path
GPT-4o-mini generates answer from retrieved context
Technical Specifications
Ingestion
- Document parsing (PDF, DOCX, MD, TXT)
- Intelligent chunking with overlap
- Binary quantized embeddings
- Duplicate detection via SHA256 hashing
Storage
- Qdrant vector database
- FalkorDB graph database
- PostgreSQL for metadata
- Redis for semantic caching
Deployment
- Docker Compose orchestration
- Next.js microfrontend architecture
- FastAPI backend with async support
- Auth0 OAuth integration
Use Cases
Knowledge Management
Centralize organizational knowledge from documents, wikis, and notes. Query naturally, get relevant context.
Research Assistant
Connect insights across multiple documents. Find relationships, trends, and contradictions automatically.
Compliance & Audit
Track document versions, access patterns, and query history. Full audit trail for regulated industries.
