-
-
working of JAX
-
calling the tools
-
copyign the code for checking
-
proposed the fix
-
applied the fix and code is ready for execution.
The Idea Behind JAX
I've always been interested in AI, and I wanted to do more than just use AI tools. I wanted to understand how they actually work and build something of my own. That's how I came up with the idea for JAX, my personal AI coding assistant.
I imagined having an assistant I could talk to while coding, one that could look at my code, help me find mistakes, and suggest ways to fix them. I wanted it to feel more like a coding buddy than just another chatbot.
How I Built It
I built JAX using Python and connected it to NVIDIA Nemotron through Nebius Token Factory. I worked on bringing together voice interaction, AI responses, and tools that let JAX interact with actual project files.
JAX can read code, inspect project files, analyze potential errors, and work with tools for testing and proposing fixes. One thing I wanted to get right was control over code changes. Instead of immediately changing a file whenever it finds a possible issue, JAX can propose a fix and wait for approval before applying it.
Challenges I Faced
Getting all the different parts to work together was one of the hardest parts. Making an AI respond to a prompt is one thing, but making it choose tools, use their results, and work with actual files is a different challenge.
I also had to think about what could go wrong when an AI suggests changing code. I wanted to make the process safer by separating a suggested fix from the actual change and adding checks around file modifications.
With the hackathon deadline approaching, I had to prioritize the core features and make practical decisions under time pressure. I used AI assistance during development, but the process also gave me a clearer idea of what I need to understand and improve next.
What I Learned
This project gave me practical experience with AI agents, APIs, language models, Python tools, and voice interaction. More importantly, it helped me understand the difference between simply asking an AI a question and building an application that can take actions through tools.
I still have a lot to learn, and I want to go back through the project and understand every part properly. I don't just want a project that works; I want to understand how it works and eventually be able to rebuild its core components myself.
What's Next for JAX
This is just the beginning. I want to make JAX better at understanding code, improve its reliability, add more useful tools, and continue making its code-editing process safer.
For me, JAX isn't just a hackathon project. It's the beginning of a longer journey into AI agents and building intelligent systems of my own.
Built With
- ai-agent
- ai-agents
- generative-ai
- llm
- nebius-token-factory
- nvidia-nemotron
- python
