MVPilot

Overview

MVPilot was built for NVIDIA's Hack-a-Claw hackathon at UCSC. The idea was to help hackathon teams move faster from idea to MVP by using AI agents to break a project down, surface blockers, plan implementation steps, and coordinate repo-aware development work.

Problem

Hackathon teams lose a lot of early hours arguing about scope, sequencing tasks, and figuring out what's actually buildable in the time they have. We wanted an agent workflow that could absorb a rough idea and quickly turn it into a realistic plan a team could execute.

My Role

I was Person 1 on a four-person team and focused on the agent, backend, and tooling side of the product. My work centered on the orchestrator, the reasoning flow, MVP validation, GitHub and project context, tool contracts, blocker detection, and memory/RAG-style context. I did not own the frontend — that was built by teammates as part of the team's product surface.

Backend / Systems Work

Built and owned the agent orchestration logic that coordinated the MVP-building workflow. Worked on the Nemotron-powered reasoning flow for validating ideas, planning MVP steps, and identifying blockers. Added repo and project context handling so the system could reason about implementation work with more relevant information, and worked on GitHub OAuth, tool contracts, and repo policy checks. Helped define how the system moved from idea intake into implementation planning.

Product Logic

MVP validation deciding whether an idea is realistically buildable in a hackathon timeline. Timeline planning that broke large ideas into practical build steps. Blocker detection that surfaced missing APIs, unclear scope, risky dependencies, and setup issues. A memory and context layer that kept useful project details available across the workflow instead of being lost between steps.

Team Process

Hackathon team project across four people. The frontend existed as part of the team's product surface, but my contribution wasn't frontend — I focused on the agent and backend systems that made the workflow actually behave like an agent loop rather than a single prompt.

Outcome

A hackathon MVP built around accelerating early-stage product development for teams. It's best positioned as an AI, backend, and product-engineering project. My main accomplishment was the orchestrator and reasoning flow that made the system behave like a real agent workflow.