Ramen: Self-Structuring Knowledge Graph Engine Aims to Replace Spreadsheets with AI-Powered Reasoning
The Tool
Ramen (Relational Agentic Memory & Emergent Network) is a new open-source project that combines LLM-powered planning with knowledge graph technology to create a "living" data structure that organizes itself as you use it.
The Problem It Solves
There's a gap between rigid enterprise systems (ERP/CRM) and the agile but fragmented world of spreadsheets — what's known as "Shadow IT." Ramen targets this gap with an inference-first architecture where the system never blocks a user because a field or relationship type doesn't exist yet.
How It Works
Key Features
- Self-structuring: Creates entity types and relationships on the fly
- Semantic deduplication: Automatically merges duplicate records (e.g., "Dreadfort Lords" → "House Bolton")
- Multi-hop traversal: Navigate complex relationship chains
- Emergent ontology: Schema evolves as data is added
- LLM-powered: Uses GPT-4o via LlamaIndex for reasoning
Tech Stack
- Language: Python
- Frontend: Streamlit
- LLM Integration: LlamaIndex + GPT-4o
- Database: Supabase (PostgreSQL + pgvector)
The Demo
The project demonstrates its capabilities with a Game of Thrones scenario:
- Add houses, characters, and military forces in natural language
- System auto-creates entity types (House, Person, Military Force)
- Merge duplicate records through semantic reasoning
- Infer relationship directions and compute aggregate military strength
Example Queries
- "What are the Lannister and its allies' combined force?"
- "Compare Stark forces vs. Lannister forces"
- "Identify traitors and their military capabilities"
Why This Matters
Ramen represents a new approach to data management:
- No upfront schema design: Start adding data immediately
- Natural language interface: Non-technical users can build complex data structures
- Graph reasoning: Relationships are first-class citizens, not afterthoughts
- Living system: The knowledge graph grows and evolves with your business
Current Limitations
The project is an architectural prototype with known limitations around permissions, scalability, and production readiness. But as a proof of concept for inference-first knowledge management, it's compelling.
Ramen is available on GitHub under open source license.