Jovia Minallah Matata

Contents

  1. Velora

    In progress
  2. FlowBoard

    In review
  3. Yosync

    Reworking
  4. ScaleUp Build

    Live
  5. Rydr

    LivePrize winner
← Contents

Chapter I

In progressSolo

Velora

An AI agent that plans big life changes, like moving abroad or starting university, as missions with tasks, deadlines and documents.

Velora dashboard listing upcoming tasks across three missions
Fig. 1. The dashboard: upcoming tasks across every mission.
Role
Everything: product, design, full stack development and the AI agent
Built with
Next.js · TypeScript · MongoDB · Auth.js · Gemini · Tailwind CSS · Playwright

The problem

Moving abroad, changing careers or starting university means dozens of linked tasks, deadlines and documents, usually scattered across notes, emails and spreadsheets. Velora started as a hackathon build. Its AI agent ran on an external cloud service, and when the credits ran out, the agent was lost with them.

What I did

  • Rebuilt Velora from the ground up with the agent running inside the app, under version control, on free tier infrastructure (Vercel, MongoDB Atlas and Cloudflare R2).
  • Designed it to be trusted: the agent can only change things through typed tools, risky actions need the user's confirmation, and every change is recorded and can be undone.
  • Locked down the API: requests without a session are refused, and requests for someone else's data get a plain "not found".
  • Made every feature work with the AI switched off, and wrote browser tests that prove it.
  • Set up CI that runs secret scanning, linting, type checks, unit and browser tests, and evaluations of the agent's answers on every change.

The result

In progress and launching shortly. Missions, tasks, the timeline and documents work today; the screenshots are from a local build with the AI switched off.

A Velora mission page for relocating from Kampala to Berlin, with tasks and recent activity
Fig. 2. A mission, with its tasks, deadlines and activity log.