DEV Community

Kaden Hunt
Kaden Hunt

Posted on

Building CODA: My AI Coding Assistant (As a Teen Dev)

Hi there!
I’m Kaden, a 16-year-old developer building an AI assistant called CODA: A Creative Optimization & Development Assistant. CODA is designed to create, edit, and optimize code by analyzing your entire project—because it actually has local access to your files. This means it can give advice that makes sense in the full context of your work, not just a single file or snippet.
I started building CODA a couple of weeks ago, and I’ll be honest—I'm not some genius. I used ChatGPT a lot to help write the code. But soon, I hope to rely on CODA alone. Right now, I'm in Phase 2, where CODA can read all files in your project, understand and generate code based on the full context, and offer targeted coding advice.
My plan for the future is to make a smart debugger, an improved memory system (because I'm currently using .json files), Text-to-speech, speech-to-text, and even a wake word so it starts feeling more like JARVIS.
If you'd like to stick around, I’ll be posting another update when Phase 2 is complete, and outlining the roadmap for Phase 3.
Thanks for reading! Let me know what you think, and if you know any ways I can optimize my progress, I'd love to hear them.

Top comments (5)

Collapse
 
anchildress1 profile image
Ashley Childress

Love the creativity! Sounds like you've put a lot of thought into it and that's exactly the kind of planning that ultimately makes things easier down the line. I'll give you a friendly challenge though - you've obviously got your bases covered with all the things it could possibly do in the future, which is great!

Have you put any thought into what it can't? Where's your limitations at? Those sorts of answers are equally impressive, if not more so, because it shows you have the ability to both think critically about the problem and that you've taken time to identify your thresholds up front. Bonus - you're on your way to fixing bugs before they actually happen, which is a skill I don't see anywhere near as often as I'd like!

Also, don't underestimate the value of a well structured json file. Especially when it comes to AI. They can be incredibly powerful in their own right, if you use it correctly 😄

Keep up the great work though! It's rare to see someone we even years older than you who sees that level of possibility and then follows through with a plan broken down into something manageable. Looking forward to your next update!

Collapse
 
drumdeedoo profile image
Kaden Hunt

Currently my limitations are CODA taking too much RAM using a decent LLM, so I hope to make hardware in the future that does the processing independent from the main CPU. That's only once I have a working AI though...

Collapse
 
anchildress1 profile image
Ashley Childress

Hey, just wanted to say - I’m definitely not a hardware expert, but I’ve been around enough code to know there’s always something to puzzle through, like your RAM issue. Even the pros hit walls where it feels like nothing makes sense and there’s no obvious answer. Trust me, you’re not alone there!

You’ve taken on a really ambitious project - seriously impressive! A lot of devs your age (and plenty older) wouldn’t even try. That said, sometimes you run into limits that just come with the territory (like hardware). Upgrading RAM would totally help, but I get that it’s not always an instant fix. So, what are you working on as a workaround right now? Is there anywhere you’re stuck or something you’re trying to optimize to use less memory?

Just to be clear: I’m not giving you a hard time at all! Figuring out what to cut or compromise is a tough skill, but it’s also what real-world dev work is like. Yeah, it stings sometimes, but it’s totally normal, and it’s how you get better. 😁

Let me know where you’re hitting a wall and maybe we can brainstorm some ideas!

Thread Thread
 
drumdeedoo profile image
Kaden Hunt

I'm using Ollama for their LLMs, but the LLM for scripting is about 13 GB of RAM. So I'm using a smaller model that uses about 5 GB to make the script with, that controlls the LLM and has a double checking system for the scripts it generates. Once I make the whole thing, I'll make hardware, and upgrade the LLM model, to improve the overall performance. Then I'll add an installer .exe so I can give out some copies to my game dev friends.

Collapse
 
drumdeedoo profile image
Kaden Hunt

Well I've hit a wall. I've realized Coda won't be very helpful to me, and though I love ai, and love curating personalities and functions for them... I don't know what to make. A GM? A Story generator? I don't know anymore.