deX AI

Self Hosted
Why deX AI?
Chat Screen
Setup
Hosting Guide

Inspiration

As a developer, I noticed how much time teams spend trying to navigate codebases, understand existing functionality, and write compatible code. I wanted to create an AI companion that not only helps understand code but also accelerates development by letting developers naturally chat with their codebase about features, bugs, and improvements.

What it does

deX AI integrates with Bitbucket repositories to analyze, understand, and index your codebase. Developers can have natural conversations about their code - from understanding architecture to finding security vulnerabilities and generating solutions for new features or bugs.

How we built it

We developed deX using Forge for Bitbucket integration using Bitbucket main menu page module and Bitbucket events The first part handles codebase indexing through AST parsing and embedding generation, while the second part implements a sophisticated retrieval system using HyDE queries and re-ranking for accurate responses. Part 1 Part 2

Privacy & Security

deX AI prioritizes data security and privacy. Choose between our secure hosted solution or self-host the entire stack. Your code is never used for AI training, and all processing happens within your control. Detailed self-hosting instructions are provided for complete data ownership. Setup

Supported Languages

Java, Python, Rust, Go, Ruby, PHP, C++, JavaScript, TypeScript, HTML, CSS, SQL, JSON,Bash, JSX, TSX

Challenges we ran into

The 25-second Forge function timeout was a problem due to long AI calls. Implementing language-specific queries across multiple programming languages and ensuring accurate code understanding while maintaining context presented significant technical challenges.

Accomplishments that we're proud of

Successfully implementing RAG for code understanding across multiple languages. The system can effectively analyze codebases and provide contextually relevant responses while maintaining privacy through optional self-hosting capabilities.

What we learned

Gained deep insights into code analysis techniques, including AST parsing, semantic search optimization, and the effective use of cross-encoders for result re-ranking. The project enhanced the understanding of building privacy-conscious AI solutions.

What's next for deX AI

I'm focusing on production readiness and seeking beta testing partnerships with companies. Future plans include expanding language support, enhancing code generation capabilities, and integerating it with Jira.