RepoQuest
A Codex Community Hack finalist that maps an unfamiliar codebase and guides a new engineer through an example contribution.
RepoQuest experiments with reducing the time companies spend onboarding new engineers by turning an unfamiliar repository into a mind map and a guided first contribution.
The situation.
When a new engineer joins a company, experienced teammates often have to spend hours explaining the codebase, its architecture, and the path to a safe first change. Documentation helps, but it can be incomplete or disconnected from the repository the engineer actually needs to understand.
I created RepoQuest during the Codex Community Hack in London to test whether part of that training could happen automatically. The idea was to inspect a repository, turn its structure into a useful mind map, and then guide the new engineer through an example contribution so they could learn by doing.
What I did.
I shaped the product concept, experience design, agent workflow, and open-source implementation during the hack. RepoQuest analyses the repository, explains how its parts connect, and turns that model into a practical onboarding path.
I did not want the system to replace the team or invent certainty. Its role is to handle repetitive orientation and prepare a useful first contribution while keeping paths, excerpts, and repository evidence available for the engineer to inspect.
How the work unfolded.
Turn the repository into a mind map
Used detector-backed analysis to identify technologies, packages, runtime roles, and relationships, then organised them into a model a new engineer could navigate.
Teach through an example contribution
Connected repository orientation to a bounded change so the user could apply what they had learned instead of stopping at a passive architecture tour.
Keep every explanation traceable
Retained file paths, line ranges, detector claims, and excerpts so the engineer could check generated guidance against the source code.
What happened.
RepoQuest became a working prototype during the Codex Community Hack on 18 July 2026 and took me into the finalist round. I presented it to a panel of founders and tech leads, explaining both the company onboarding problem and the mind map plus guided-contribution approach.
The result also won me one year of GPT Pro 20x access. That prize helped define the next stage of my builder journey by giving me sustained access to the agentic tools I used to keep experimenting and shipping.
What I learned.
A map should lead to action
Understanding a repository becomes more useful when the engineer can immediately apply that model through a small, bounded contribution.
Automation should give mentors leverage
The useful goal is not removing experienced engineers from onboarding. It is reducing repetitive explanation so their time can go toward context, judgement, and feedback.
Generated guidance needs receipts
A mind map or explanation is only trustworthy when the engineer can follow it back to the relevant files and source evidence.
A hackathon can change what comes next
The value went beyond the prototype. Presenting as a finalist and receiving a year of GPT Pro 20x access changed the tools available to me and shaped the next stage of my builder journey.
Related links.
Repositories, demos, event pages, posts, and other places connected to this project.