GBTutor

AI tutors for college courses with access to current material, and peer tutor matching where students who need help can find private, peer tutors and qualified students can be tutors and earn income.

Comment

Inspiration

Office hours and Discord queues move too slowly when you're stuck the night before an exam, so we wanted instant AI help that still hands off to a real Berkeley student who's already aced the class. There can be friction particularly in the first year with freshman and transfer as they adjust to the rigor of schools like Berkeley. This platform is inspired by those students.

What it does

GBTutor pairs an always-on AI Socratic tutor for your exact course with a marketplace of B+-or-higher verified peer tutors, letting you escalate from AI to human when you need one.

How I built it

I built a Next.js platform with a Claude-powered tutor, Transformers.js embeddings for matching, a real Fetch.ai uAgent (reasoning via ASI:One) to broker matches, Browserbase for live course-page grounding, and Arize for observability, all wired through clean adapter seams with safe fallbacks, with Stripe checkout scaffolded for payments next. Sponsor tools used:

  1. Anthropic/Claude — the Socratic AI tutor calls the real Claude API.
  2. Fetch.ai — the broker_agent.py is a real uAgent registering on Agentverse, using ASI:One for reasoning.
  3. Browserbase — creates a real session via their REST API.
  4. Arize — logs real records to their REST endpoint.

Challenges I ran into

Most of my last hour went to environment friction rather than logic such as mismatched Python versions across pip/pip3, a uagents library incompatibility with Python 3.14's stricter asyncio behavior, and getting Fetch.ai's real broker agent talking to ASI:One instead of just mocking the integration.

Accomplishments that I'm proud of

I shipped a genuinely working dual-mode tutoring flow. AI chat and peer escalation with four sponsor-tool integrations actually wired to live APIs rather than placeholder calls under hackathon time pressure.

What I learned

Verifying the exact request/response shape of each sponsor API before wiring code (and being honest when something was still a fallback or not yet complete, like Stripe) saved far more time than it cost.

What's next for GBTutor

Next is finishing real Stripe Checkout end-to-end, a real database and auth in place of the JSON dev store, Stripe Connect for tutor payouts, and completing the Browserbase integration so it extracts live course content into the tutor's context instead of just proving the session connects.

Built With

Share this project:

Updates