We've wanted to build Ari for a long time. Way before it had a name. Way before we had the money.

When we built Pinned AI Tutor, we were proud of it. It worked. Students learned. But here's the thing we never said out loud: it was a workaround. A clever one — using ChatGPT and Gemini's pinned sessions with our curated reference materials — but still a workaround. The AI couldn't remember the student. It couldn't track progress. Every session started from zero.

We knew what we wanted to build. A real AI mentor that stores conversations, analyzes them, builds a profile of how each student learns, and adapts every single time. But that costs money. Every API call, every stored conversation — that's compute. And for a nonprofit running on donations, compute was the wall we kept hitting. So we designed the system in our heads and waited.

Then we applied for AWS Promotional Credits for Social Impact. We didn't overthink the application. We just told the truth — here's who we are, here's who we serve, here's what we want to build, and here's why we can't. They said yes. $10,000 in AWS credits. That changed everything.

We went straight to teens. Why? Because they're going to live with AI longer than anyone, and most AI education for that age group is either patronizing or boring. We'd been watching teens use AI tools with zero guidance — they're already using it, just without understanding what's happening behind the screen. We figured: what if we gave them someone who could teach them properly, in a way that didn't feel like school?

We went deep into educational research — what attributes actually matter when tracking student learning? Confidence levels, knowledge gaps, learning styles, misconceptions. We built an extractor pipeline that analyzes every conversation from multiple angles. Not just "did they get the answer right," but "how do they learn best? What analogies land? Where do they get stuck and why?" Every session makes the system understand the student a little better.

The name? AI Bridge Foundation — AIBRI — Ari. It just happened. Someone shortened it, it sounded right — short, friendly, not trying too hard — and it stuck.

We added something we're calling "vibes" — six different personalities students can pick. Older Sibling, Chill, Hype Me Up, Coach Mode, Patient & Gentle, Talk Like Me. It's a small feature technically, but for teens it made so much sense. The difference between a coach voice and a chill voice is the difference between a kid coming back tomorrow or closing the tab forever. It's honestly one of the most fun things we built.

Under the hood: Amazon Bedrock with Claude Sonnet 4.6, conversations stored in S3, prompt caching so the AI doesn't lose context, an async extraction pipeline that updates the learner profile after each conversation, and progress tracking across 11 topics. None of this was possible six months ago — not because the technology didn't exist, but because we couldn't pay for it.

Ari isn't a chatbot. It's the product we always wanted to build — a fully integrated AI mentor that understands each student as an individual. Now it's live. It's free. And we're just getting started.