Docomate AI
AI documentation generator.
Created on 22nd April 2025
•
Docomate AI
AI documentation generator.
The problem Docomate AI solves
We cannot deploy our each project, but we can at least push it to github. And if you want to showcase these projects to your friends or recruiter, then in that case README is must. But creating a readme consumes a lot of time, and sometime we do leave some necessary sections to add in it. Here the Docomate AI comes, your go-to tool to generate Readmes or to understand codebase through our Chat with Codebase feature. With Docomate AI, with just the repository Url, you can create some most stunning and detailed README in just 6-8 secs and without any prompt or description.
Challenges I ran into
-
I had no prior experience building a Retrieval-Augmented Generation (RAG) pipeline. While I was familiar with using AI APIs, ingesting an entire repo posed a serious challenge. Pasting the entire codebase into a prompt works only for very small projects.
-
To address this, I built a scraper that clones the repo and compiles all its content—along with file metadata—into a single text file. I then used Langchain to chunk this file and stored the chunks and their vector embeddings in a MongoDB database.
-
The next major challenge involved MongoDB’s vector search capabilities. It turned out that MongoDB’s native vector search wasn’t compatible with the data model I used. So, I implemented a custom cosine similarity algorithm, tweaked it to return previous and next chunks for better context, and integrated it into the system. This was one of the most complex features I developed.
Tracks Applied (1)
Groq track
Groq