https://github.com/ivanvanderbyl/agentic
A minimalist Go implementation of Microsoft's GraphRAG
https://github.com/ivanvanderbyl/agentic
graphrag openai rag
Last synced: about 2 months ago
JSON representation
A minimalist Go implementation of Microsoft's GraphRAG
- Host: GitHub
- URL: https://github.com/ivanvanderbyl/agentic
- Owner: ivanvanderbyl
- License: mit
- Created: 2024-07-07T05:31:42.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-08-25T03:47:49.000Z (about 1 year ago)
- Last Synced: 2025-04-10T15:46:17.953Z (6 months ago)
- Topics: graphrag, openai, rag
- Language: Go
- Homepage:
- Size: 77.1 KB
- Stars: 15
- Watchers: 2
- Forks: 1
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# `agentic`
This package serves as a workspace for exploring state of the art RAG approaches in Go. The initial focus is on the GraphRAG approach from Microsoft, but other approaches may be explored in the future.
## Areas of Exploration
- [x] GraphRAG
- [ ] dsRAG
## Findings
### GraphRAG
GraphRAG is seems to be mostly marketing spin from Microsoft. There are some legitimately cool ideas in there such as using LLMs to extract graph relationships from a given text.
The heavy use of multi-stage summarisation, however, produces results that are not accurate to the original text or useful for an executive or research audience because they too general and lacking concrete details.
### dsRAG
Source: https://github.com/D-Star-AI/dsRAG
There are three key methods used to improve performance over vanilla RAG systems:
Semantic sectioning
AutoContext
Relevant Segment Extraction (RSE)
We will implement these in Go to explore them further.