Funding
Funding not disclosed
Founders
Product
Problem
Deploying retrieval-augmented generation (RAG) systems, especially on local or resource-constrained hardware, often results in context relevance issues and increased LLM hallucinations. Traditional semantic search methods can fail to capture critical relational information, leading to suboptimal performance in complex domains.
Solution
AGI Dreams enhances RAG systems through a graph-driven knowledge base pipeline, integrating semantic and relational retrieval for improved context accuracy. This approach leverages a "community nested" relational graph to facilitate both top-down and bottom-up information retrieval, ensuring that the LLM receives more comprehensive and relevant context. The system's visualization tools, built with NextJS and ThreeJS, allow developers to inspect knowledge graph coherence, troubleshoot clustering, and verify the context provided to the LLM. This transparency and enhanced context significantly reduce factual inaccuracies and hallucinations, particularly in specialized technical domains.
Target Audience
The primary target audience includes developers and organizations focused on deploying RAG systems on local hardware, particularly those working with small LLMs or in domains requiring high factual accuracy and minimal hallucinations.
Features
- Graph-driven knowledge base pipeline for hybrid retrieval (semantic and relational)
- Visualization tools for knowledge graph inspection and debugging using NextJS/ThreeJS
- Support for local and constrained hardware deployments of RAG systems
- Techniques to improve context relevance and reduce LLM hallucinations
- Integration of community-nested relational graphs for enhanced information retrieval
- CLI-first interfaces and native bindings for lightweight inference engines
- Configuration flexibility for chunking, model swapping, and embedding choices
- Metric monitoring for pipeline performance and RAG system evaluation