I built an AI that detects chart patterns to fight my own confirmation bias I kept losing money on trades because I'd "see" patterns that weren't there. Classic confirmation bias — when you're already in a position, your brain lies to you.

So I built a tool: upload any chart screenshot, get pattern detection in seconds.

Why Gemini over GPT-4V? Tested both. Gemini 1.5 Flash is: - Faster (~2s vs ~5s) - Cheaper (~$0.0001 per analysis) - More consistent structured outputs for this specific task GPT-4V sometimes gave me essays. Gemini stayed focused.

Stack: Next.js 14, Supabase (auth + postgres), Stripe, Vercel

The hard parts weren't the AI: - Auth edge cases (email confirmation flows, session refresh) - Credit system (what happens when API fails mid-request? refund?) - Making it not look like a hackathon project (I'm a backend guy, CSS is pain)

Is it always right? No. TA itself is debatable. But it's a second opinion that doesn't care what positions I'm holding. That's the value.

https://trinith-ai.vercel.app

Would love feedback from: - Traders who can tell me if the output format is useful - Anyone who's built with vision APIs (optimization tips?) - Skeptics who think this is dumb (genuinely want to hear why)