Inspiration Have you finished a long 8hrs of coding and your manager says he want documentations, NOW?! Or, have you read another colleague or fellow teammate's documentation and got nothing but big piece of hot mess? Many coders get obstructed in familiarizing with the codebase becuase of poorly written docs. Like the non-technical users say, I want AI to do my chores and let me do art. Well, coders' art is the code, and the documentation is the chore. Now we are bringing an AI automated documentation tool to you!
Many companies have poorly structured or incomplete documentation for their codebases, not to mention, uploading your codebase to Chatgpt will not get the job done, but get you fired!!! We are making this AI tool entirely local so documentation can be made easy for developers—especially newcomers—to understand and maintain projects. We wanted to create a solution that could automatically analyze code and generate well-organized, systematic documentation.
What It Does CodeScribe takes in source code, processes it using a trained model, and generates clear, structured documentation in markdown format. This helps developers quickly understand the functionality and structure of the code without manually writing extensive documentation.
How We Built It We designed a pipeline where the frontend sends a URL to the backend via an API. The backend fetches the content, processes it, and trains a machine learning model to analyze the code. The output is a structured .md file, which is then converted into JSON and returned to the frontend for easy display.
Challenges We Ran Into Our team has expertise exclusively in single field like backend, frountend and AI, so integrating the two was a learning curve. Finding an LLM model that could effectively process and generate high-quality documentation from code that can be fine-tuned with limited computing power and able to run locally was a significant computing challenge. Parsing and structuring complex code while maintaining clarity in the generated documentation required experimentation and refinement. Accomplishments That We're Proud Of Successfully integrating the frontend and backend despite our initial lack of experience. Implementing an automated pipeline that converts raw code into structured documentation. Overcoming challenges in training the LLM model to improve the quality of generated documentation.
What We Learned How to work with APIs to connect the frontend and backend. The importance of choosing the right machine learning model for processing code. Strategies for structuring and formatting auto-generated documentation effectively. What's Next for CodeScribe We plan to enhance CodeScribe by:
Improving the accuracy and readability of the generated documentation. Supporting more programming languages and frameworks. Adding a user-friendly interface to allow developers to customize the output. Exploring fine-tuning techniques to make the LLM model more effective for documentation generation.
Log in or sign up for Devpost to join the conversation.