Restaurant Recommendation System

Architecture documentation for a Restaurant Recommendation System.

Status

🚧 Under Development

---

v2 Architecture — Knowledge Graph + Graph-RAG + GNN + AI Agent

The v2 architecture is built on FalkorDB (Knowledge Graph), CocoIndex, PyTorch Geometric (GNN), and GPT-4o (AI Agent). It replaces traditional collaborative/content-based filtering with a graph-aware, hallucination-free chatbot experience.

Architecture Docs

Document	Description
System Overview	3-flow architecture diagram, component layout, latency analysis
Tech Stack	FalkorDB, CocoIndex, PyG, GPT-4o, LangChain — why each was chosen
Data Flow	Sequence diagrams for all 3 system flows
Graph-RAG Pipeline	Deep dive into the Graph-RAG retrieval and prompt design

Module Docs

Document	Description
01 — Data Layer	PostgreSQL schemas, CocoIndex CDC pipeline, data quality
02 — Knowledge Graph	CRKG design, FalkorDB config, Cypher query patterns
03 — GNN Engine	GCN architecture, multi-task loss, training pipeline
04 — AI Agent	AI Waiter chatbot, intent classification, LangChain integration
05 — Serving API	FastAPI endpoints, caching strategy, WebSocket support
06 — Observability	Metrics, ML monitoring, alerting, Grafana dashboards

Schema Docs

Document	Description
CRKG Graph Schema	Complete node/edge schema for the Knowledge Graph
Recipe Catalog	Recipe JSON schema, nutrition facts, NDS formula
User Profile	User profile JSON schema, segments, health priority score
Nutrition Tags	Full tag taxonomy, thresholds, LLM extraction process

API Docs

Document	Description
API Specification	All endpoints, request/response examples, WebSocket protocol, error codes