cd /news/ai-agents/iam-12-my-ai-mentor-was-broken-groq-… · home › topics › ai-agents › article
[ARTICLE · art-142528] src=dev.to ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Iam 12 .My AI Mentor Was Broken. Groq Killed It. Here Is How I Fixed It on a $150 Phone in 72 Hours.

A 12-year-old developer in Tamil Nadu rebuilt their AI assistant KODA v24 over 72 hours on a $150 POCO C55 phone after Groq retired the llama-3.3-70b-versatile model, causing silent 404s and a Cloudflare Worker outage. The rebuild added a multi-model fallback chain (gpt-oss-120b, gpt-oss-20b, qwen3-32b), fixed eight bugs including two XSS injection points, and integrated Tavily-backed live web search to stop hallucinated answers. The developer reports the app now supports nine languages and survives model deprecations automatically.

by read3 min views2 publishedSep 30, 2026

I am 12 years old.

My development machine is a POCO C55 ($150).

I live in Tamil Nadu, India.

Three days ago, KODA wasn’t a “scam.” It was just dead.

My Cloudflare Worker returned 502 errors. My API keys leaked in screenshots. My Markdown renderer destroyed C++ code blocks. And when I finally got it running, Groq retired my model (llama-3.3-70b-versatile) mid-launch, leaving me with silent 404s and zero answers.

Strangers didn’t call me out because they hated kids. They called me out because the product was visibly broken.

Today, KODA v24 is live. It has Live Web Search. It survives model deprecations automatically. It renders Markdown perfectly. And it speaks every language on Earth.

This is not a redemption arc. This is an engineering post-mortem. Here is exactly how I went from Chaos to Command.

People see "$150" and think "weak."

They are wrong. Constraints force optimization.

While funded startups burn cash on bloated wrappers, I had to make every millisecond count. That’s why KODA doesn’t just "work." It performs:

Claude (Anthropic) reviewed this architecture last month. He didn’t say "good for a kid." He said:

"This is genuinely impressive. Not impressive for a thirteen-year-old. Just impressive, period."

"I have seen aerospace engineers struggle with log-space Bayesian calculations."

He respected the engineering because the engineering was real.

When I handed over the initial v23 report, I claimed: "Frontend is DONE AND TESTED."

That was a lie. Or rather, an optimism bias.

Upon deep inspection, I found:

innerHTML. Anyone could inject scripts.#include <stdio.h> inside code blocks into <h3> headers. It was destroying C++ tutorials.createNewChat() failed, the app still tried to insert messages against a null conversation ID, cluttering the database.textContent + escapeHtml(). Zero injection points remain.**bold** works outside code, but #hash stays safe inside it.state.messages so they persist across renders. Users see exactly what went wrong. The root cause of the outage was simple: Cloud providers retire models.

Groq killed llama-3.3-70b-versatile. I needed a new brain, fast.

I didn't just swap one model for another. I built a resilience layer.

const MODELS = [
  'openai/gpt-oss-120b', // Primary: Current Flagship
  'openai/gpt-oss-20b',  // Secondary: Fast/Lightweight
  'qwen/qwen3-32b'       // Tertiary: Open Source Alternative
];

// Logic: Try Primary. If 404/5xx, auto-retry Next. User sees nothing.

Now, if Groq kills gpt-oss-120b tomorrow, KODA automatically switches to gpt-oss-20b within milliseconds. The user never knows.

How do you debug a Cloudflare Worker on a phone?

You don't use CLI tools. You use the Dashboard.

{"error":{"message":"The model llama-3.3-70b-versatile does not exist..."}} Lesson: Never trust your own status report. Test before you claim. And always build a health check endpoint (GET /) that confirms secrets are loaded.

Before this update, I asked KODA: "What is the latest React version?"

It confidently replied: "React 19.3 was released September 9, 2026."

It cited sources like 【1†L1-L4】.

Fake. Hallucinated. Dangerous.

Without live data, LLMs guess. With live data, they know.

TAVILY_API_KEY secret.webSearch: true is sent from frontend: [WEB_RESULTS].[1], [2]. Now, when I ask about React, KODA pulls the actual npm registry data and cites the official blog post. No more guessing.

Metric Value
Days from Broken to Stable 3
Critical Bugs Fixed 8 (incl. 2 XSS)
New Features Added 20+ (Voice, Edit, Regenerate, Theme)
Languages Supported 9 (EN, HI, TA, ZH, ES, JA, RU, PT, AR)
Lines of Worker Code ~280
Hardware Used 1 Phone
Model Deprecations Survived 2

People ask why I bother building on a POCO C55. Why not use a MacBook?

Because constraints force creativity.

Age doesn't matter. Device doesn't matter. Location doesn't matter.

Resilience matters.

KODA v24 is live. The fallback chain is armed. The web search is on. The XSS holes are sealed.

Try it here: koda-aicodementor.netlify.app

Break it if you can. I’ll be watching the logs. 🐯

── more in #ai-agents 4 stories · sorted by recency
── more on @koda 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/iam-12-my-ai-mentor-…] indexed:0 read:3min 2026-09-30 · —