Leet++

Inspiration

I'm preparing for software engineering interviews, and the whiteboard portion is the hardest to practice alone. The AI coding tools I've used work with typed code, but interviews and tutoring often happen with a marker, out loud. I wanted a practice partner that could watch the board and respond like an interviewer, without giving away the answer.

What it does

Leet++ is a live problem-solving assistant for the whiteboard. The goal: point a camera at your board, pick a problem, and solve it out loud while an AI reads your handwritten code and talks with you in real time. It asks questions and flags logic errors, gives hints in stages, and does not write code for you. Logic and approach errors take priority; syntax slips are noted but not interrupted.

How we built it

Planned stack:

• Next.js frontend for problem selection, camera, and microphone
• Perceptron Mk1.5 for reading handwriting and board layout
• FastAPI backend with rule-based logic deciding when the interviewer speaks
• Gemini Live API for real-time voice

Challenges we ran into

Our original voice provider's hackathon credits were withdrawn mid-event, so we redesigned around the Gemini Live API. The main design problem so far has been timing: deciding when an interviewer should interrupt rather than just what it should say.

What's next for Leet++

Spaced-repetition practice by skill, running code against tests, and expanding to tutoring for any subject worked out by hand.

Built With

• claude
• fastapi
• gemini
• next.js
• perceptron