Show HN: KnowLang – An open-source tool for understanding complex codebases

1 point

a year ago

Hey HN, After two years in big tech while side hustling after work hours, I've found that AI coding assistants like Copilot and Cursor have significantly boosted my side project development speed. However, these assistants haven't been as helpful for my regular work, where I spend at least 50% of my time reading and analyzing codebases and inter-repository dependencies.

This experience led me to develop KnowLang (https://github.com/kimgb415/know-lang), an open-source tool aimed at helping engineers gain a comprehensive understanding of enterprise-level codebases.

The current version of KnowLang follows a straightforward technical approach:

1. Parse the code

2. Summarize the code

3. Index both code & summary into a vector database

4. Provide a RAG (Retrieval-Augmented Generation) chatbot interface

While this may not seem groundbreaking compared to existing RAG systems, I plan to focus on the following enhancements:

1. Developing code-specific RAG systems with code-focused reranking and embeddings (e.g., voyage-code-3)

2. Implementing an automatic LLM fine-tuning system to reduce sole reliance on RAG

3. Enabling inter-repository knowledge awareness through multiple depth/layers of RAG

I've published a blog post detailing KnowLang's architecture and performance on the Hugging Face blog: https://huggingface.co/blog/gabykim/knowlang-first-demo

You can try out KnowLang via:

- demo Hugging Face Space: https://huggingface.co/spaces/gabykim/KnowLang_Transformers_...

- Installing the package: `pip install knowlang`

- Exploring the GitHub repository: https://github.com/kimgb415/know-lang

I would love to hear your thoughts, feedback, and suggestions. Feel free to ask any questions – I'll be around to discuss and learn from the community.

Looking forward to your insights!

Gaby