Graveyard

Builds from 24-hour, 5-hour and one-hour sprints. Some stalled, some got pitched. The only thing truly buried here is a sleep schedule.

Flowers are counted on your device only. Leave as many as you like.

  1. ApprovID

    April 2026
    XRPL Commons Hackathon, Paris

    “Forced blockchain, 6 hours. Survived by a better idea.”

    Status Pitched at Paris Blockchain Week after the event.

    Read the inscription

    What we tried. Our first ideas used a blockchain because the theme said so, not because the chain added anything. About six hours and two pivots in, we landed on private attribute verification, where it does.

    What I did. Challenged the weak early directions and helped align the team on the stronger use case. Spent most of the build on Next.js, API routes, the ZKPassport SDK integration and embeddable verification flows. Focused on making verification easy for a business to drop into its own checkout.

    The team. Explored and pivoted together, then shipped a working prototype inside 24 hours.

    Afterwards. I later pitched ApprovID to founders and ecosystem teams at Paris Blockchain Week and kept building relationships in the industry afterwards.

    Full story

    for the two pivots that came before it

  2. AidAtlas

    April 2026
    Claude Builder Club @ Trinity hackathon

    “Mapped the need. Asked the next question.”

    Status Working MVP at the event. What happened next is still to be written up.

    Read the inscription

    What we tried. Knowing where need exists is not enough. The harder question is where constrained help should go first, and how visibility turns into coordinated action.

    What I did. Came up with AidAtlas and defined the problem and product direction. Prioritised the MVP, kept the team moving, and contributed to the implementation.

    The team. Built the platform together with React, TypeScript, the Claude API and Convex. Combined humanitarian indicators from INFORM, Aqueduct and JMP, starting with water-related need.

    Full story

    for the volunteers it was built for

  3. UnderTheRadar

    April 2026
    Give(a)Go × Daytona hackathon

    “The sandboxes said no. We shipped anyway.”

    Status Working hackathon MVP. Current status to be confirmed.

    Read the inscription

    What we tried. Opportunities in Ireland are fragmented across Eventbrite, Meetup, Luma and a dozen group chats. You only hear about the good ones afterwards.

    What I did. Originated the idea and stayed involved from problem definition through prioritisation, implementation and delivery. Helped coordinate the team, including a late pivot when the planned architecture stopped working. Worked across the PostgreSQL-backed API, source ingestion, ranking, search, preference matching and the React frontend, alongside teammates.

    The team. Built the ingestion pipelines and a React/TypeScript app with a Hono API, PostgreSQL and Drizzle ORM. Dropped the multi-sandbox Daytona design late, after network restrictions on the available tiers blocked it.

    Full story

    for the multi-sandbox architecture

  4. AI Grand Prix

    August 2026
    OpenAI × Formula Trinity AI Grand Prix

    “Here lies the stable driver. Swapped out at the deadline.”

    Status Best practice lap 27.84 s. Official placement not recorded.

    Read the inscription

    What we tried. Use LiDAR readings to choose steering and speed, fast enough to matter, in about an hour.

    What I did. Wrote an aggressive, wall-assisted racing-line controller in Python and tuned it to a 27.84-second best practice lap. Then, moments before the deadline, submitted the experimental version instead of the stable one.

    Afterwards. The lesson stuck: protect the known-good build. I have since been reading into reinforcement-learning approaches.

    Full story

    for the experimental build that went in instead

  5. TeamFlow Agent

    April 2026
    Give(a)Go Zero to Agent hackathon, with Vercel and Lua AI

    “Knew who was blocked before they did.”

    Status Team prototype. Current status to be confirmed.

    Read the inscription

    What we tried. Under pressure, teams lose track of who is doing what, leave blockers unspoken and quietly duplicate work.

    What I did. Took part in the team build. A breakdown of my own part is still to be written up.

    The team. Built a context-collecting VS Code extension and a central backend. Built a coordinating agent, a Vercel-deployed dashboard, a chatbot on the Vercel AI SDK and an MCP server.

    Full story

    for every unspoken blocker

  6. Craicdown

    April 2026
    Craicathon

    “Five hours. Five people. One stern push notification.”

    Status Pitched at the event. Current status to be confirmed.

    Read the inscription

    What we tried. Mindless scrolling is passive. Could a few seconds of Irish make it a moment of awareness instead?

    What I did. Part of a five-person team that went from idea to MVP to pitch in about five hours.

    The team. Designed the awareness loop, built the mobile MVP and pitched it, and ran into mobile OS permission limits on the way.

    Afterwards. Takeaway: for a five-hour build, the web is often faster than mobile.

    Full story

    for the OS permissions dialog

Got a hackathon that needs one more teammate?

Recruit me