Effortlessly generate GitHub READMEs or interact with your codebase using AI.

Build a creative and interactive multimodal application powered by Groq, leveraging the capabilites to solve real-world problems with a focus on user experience and innovation.

Writing code and debugging are the most exciting aspects of building any tech product. However, not every developer excels at writing documentation. Many developers struggle with writing clear and concise README files.

Docomate AI aims to solve this by allowing developers to effortlessly generate READMEs for their public GitHub repositories. There are no complex prompts or tedious forms—just paste your repo link, select the sections you want, and the README is generated in seconds.

Another major challenge, especially for beginners exploring open-source projects, is understanding the structure of an unfamiliar codebase. With Docomate AI, users can chat with the codebase, ask questions, and get contextual insights, making navigation much easier.

Currently, Docomate AI achieves around 55% - 65% accuracy for small to mid-scale projects and approximately 80% - 85% accuracy for micro-scale projects (repos with 4–7 files).

Gitanshu Sankhla (Solo Hacker so No Team)

• Gitanshu Sankhla - Responsible for Frontend, Backend, Database, and Generative AI aspects of the project.

  

• As a regular GitHub user, I often push even my smallest toy projects to the platform. However, I can’t always deploy them, so when I share these repositories, people have no clue how to set them up, what the project does, or what its features are. Every project deserves a good README, but writing one can take an entire weekend.

• That’s when I thought—what if we could generate READMEs in just one click and under 8 seconds?

• 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.

• Initially, I wanted to only focus on generating README files from a GitHub repo link. However, during the hackathon, I realized that understanding someone else’s codebase is a much bigger problem for beginners. This led to a major pivot—I decided to implement a chat interface where users could chat with the codebase to understand it better.

• While experimenting with MongoDB’s vector search capabilities, I discovered that it didn’t suit my specific data structure. This unexpected limitation forced me to brainstorm an alternative. After extensive research and testing, I ended up building a custom cosine similarity search algorithm, which actually performed better in my use case.

• I also faced a breakthrough moment while designing the scraping logic. Initially, I tried directly feeding code into the LLM, but it wasn’t scalable. The idea to build a custom repo scraper that parses file structure and contents with metadata into a vectorized knowledge base allowed me to scale the application to larger projects.

• Groq: I utilized chat completion endpoint for generating READMEs and enabling chat with codebase.

Most important features of Docomate AI are:

• ✅ Authentication and authorization for secure access to user resources.
• ✅ Users can create projects, with each project supporting multiple READMEs, documentation files, and chats.
• ✅ Built-in Markdown editor to edit and download READMEs at any time.
• ✅ User can save and download their README's
• ✅ Profile page where user can view and update their profile.
• ✅ App settings, with settings user can change their app theme or reset their account password.
• ✅ Chat with your codebase — users can ask questions and explore unfamiliar repositories with ease.

• Demo Video Link: Youtube Video
• Pitch Deck / PPT Link: Google Slides PPT

• All members of the team completed the mandatory task - Followed at least 2 of our social channels and filled the form (Details in Participant Manual)
• All members of the team completed Bonus Task 1 - Sharing of Badges and filled the form (2 points) (Details in Participant Manual)
• All members of the team completed Bonus Task 2 - Signing up for Sprint.dev and filled the form (3 points) (Details in Participant Manual)

• Node JS, Nest JS and Typescript.

• API KEYS

  API Credentials	Website
  MongoDb Credentials	MongoDb Atlas
  SendGrid Credentials	Sendgrid Mail Service
  Jina Embeddingd API	Jina v2 Embedding Modal
  Groq API	Groq Console

• To configure environment variables, navigate to the backend directory and copy .env.development.sample to .env.development. Then, paste your API keys from the above links.

Better prefer Backend Repository for better instructions

# Clone the backend repository
git clone https://github.com/Docomate-ai/docomate-ai-backend

# Install the necessary dependencies
cd docomate-ai-backend
npm install

# Copy the .env file and put the credentials
cp .env.development.sample .env.development

# Start the development server
npm run start:dev

Better prefer Frontend Repository for better instructions

# Clone the frontend repository
git clone https://github.com/Docomate-ai/docomate-ai-frontend

# Install the necessary dependencies
cd docomate-ai-frontend
npm install

# Start the development server
npm run dev

Docomate AI v1.0.0 successfully addresses two major developer pain points: generating GitHub documentation (README) and understanding complex codebases through chatting with it. In future iterations, I plan to introduce the following features:

• Bring your own Model: Users with Groq pro subscriptions can use their own API keys and select the model of their choice. (Expected Release: 30th April 2025)
• Prompt Customization and Section Addition: Users will be able to customize prompts and add new sections in the README. (Expected Release: Mid May)

• Documentation Site Generation: Generate full documentation sites for large-scale projects.
• OpenAPI docs generation: Auto-generate API documentation following OpenAPI standards using Groq.
• Talk to codebase in native language: Multilingual support for talk with codebases.

Participating in Hackhazards'25 has been an incredible journey. I truly stepped out of my comfort zone to create this powerful project product. I believe it has the potential to positively impact many developers.

Huge thanks to the Namespacecomm community and @pradeepto sarkar for organizing this amazing event. This was my first application integrating Generative AI model and a full RAG pipeline.

Thank you!