Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Completed research on semantic retrieval augmented generation through novel semantic similarity graph traversal algorithms.
| Date | Stars |
|---|---|
| 2026-07-31 | 271 |
| 2026-08-04 | 271 |
| 2026-08-06 | 271 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<h1 align="center">Novel Semantic Similarity Graph Traversal Algorithms for Semantic Retrieval Augmented Generation Systems</h1>
---
<div align="center">
<img src="docs/3D_VIS_1.png" width="49%" />
<img src="docs/3D_VIS_2.png" width="49%" />
</div>
<h1 align="center">Abstract</h1>
<div align="center">Semantic retrieval augmented generation (RAG) systems are designed to provide large language models (LLMs) with the necessary context to answer user queries precisely and accurately from a database or semantic similarity graph. However, the innate challenge with traditional RAG systems is their dependence on fine-tuned, raw text matching via a vector store based on the user's query. This research proposes novel traversal algorithms that meaningfully move through knowledge bases with the goal of extracting highly relevant and precise information for user queries. These traversal algorithms provide the foundational bedrock in the pursuit of accurate semantic RAG systems and can be easily built upon or fine-tuned depending on use case. </div>
---
<p align="center">
<a href="https://www.youtube.com/watch?v=HcHFBAQqe_k">
<img src="https://img.youtube.com/vi/HcHFBAQqe_k/maxresdefault.jpg" alt="Video" width="600">
</a>
<br>
<a href="https://www.youtube.com/watch?v=HcHFBAQqe_k">
<img src="https://img.shields.io/badge/▶%20Watch%20Video-red?style=for-the-badge">
</a>
</p>
---
<h1 align="center">Foreword</h1>
This `README.md` contains a concise publication of the research discovered in this repository. For a thorough, guided demonstration with interactive elements (algorithm testing, semantic similarity graph visualizations like the ones above), please view the Jupyter notebook in the root of this repository. It is intended to be run locally via a Python `venv` due to dependencies.
> **Terminology update (2025-01):** Earlier drafts described this work as "knowledge graph" research. After community feedback we now refer to the data structure as a *semantic similarity graph* (SSG). An SSG stores untyped nodes connected by cosine-similarity edges; it does not provide ontological typing, relation semantics, or rule-based inference the way a formal knowledge graph does. Throughout this document, "semantic similarity graph" and "SSG" are now used to accurately describe the system.
To run the notebook, first ensure you have Python 3.12 and `ollama` installed, then run:
```commandline
git clone https://github.com/glacier-creative-git/semantic-similarity-graph-traversal-semantic-rag-research
cd semantic-similarity-graph-traversal-semantic-rag-research
python3.12 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
jupyter notebook research_demonstration.ipynb
```
Additionally, if you are interested in running the full pipeline, you may do so. First, set your environment variables in the `.env` file.
```commandline
# OpenAI Configuration
OPENAI_API_KEY=openai-api-key
# Ollama Configuration
OLLAMA_BASE_URL=http://localhost:11434
# Anthropic Configuration
ANTHROPIC_API_KEY=anthropic-api-key
# OpenRouter Configuration
OPENROUTER_API_KEY=openrouter-api-key
# DeepEval Platform Configuration
CONFIDENT_AI_API_KEY=deepeval-api-key
```
You'll need to set up an account with DeepEval to get your `CONFIDENT_AI_API_KEY`. You'll also need to get a dataset added through the dashboard manually if this is your first time using DeepEval. It is *strongly recommended* you use `1qa-intradoc-reasoning-multicontext.csv` located in the `datasets` directory for your convenience.
Then take a look at the `config.yaml` which holds the configuration settings of the entire repository. Upon publication, it will be configured to test *all seven algorithms* on a single DeepEval golden. To do this, run:
```commandline
python benchmark.py --help
```
Lastly, two notes. First, reranking code exists in the repository but has been depreciated. The final Excerpt of 35,436 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:7f416025ef0fb025, desc:retrieval-augmented generation, desc:retrieval augmented