Skip to content
Chukwudum Chukwuedo

Lab

Lab

A small number of things I have built, am testing, or have deliberately not released. The live build is in front of you. Shipped applications link out to the running product. The rest are research threads, labelled honestly and kept quiet until they pass an audit.

Live now

Live build

Elevatech Connection Compass

A live, on-device matching tool built for a 30-minute breakout and demonstrated on stage.

Capabilities
Product design · Frontend engineering · On-device matching
Evidence
public
Disclosure
general

System view

Architecture and relationships

Diagram not yet published

The Connection Compass is a client-side guided flow: answers stay on the device, matching runs in the browser, and the result is a ranked shortlist with reasons.

The problem

A conference programme is a poor interface for a room full of people with different reasons for being there. The printed grid tells you what is on. It does not tell you which sessions connect to your work, which conversations are worth interrupting, or where the useful collisions are likely to happen.

The Connection Compass was built for a 30-minute breakout at Elevatech Summit London. The brief was not to describe personalisation. It was to put a working tool in front of the room and let people use it.

Intended user

The first user is an attendee who wants a short, useful shortlist without filling in a long profile or handing over a contact list. The second user is the person running the room, who needs the tool to work on a shared network, on whatever device is in someone’s hand, without a support team.

Architecture and interface

The build is a small client-side application. The attendee answers a short set of questions, the matching runs in the browser, and the result is a ranked set of sessions and people to seek out. The matching logic is deliberately modest: a transparent scoring model is easier to explain, inspect and adjust in the room than a larger model that nobody can interrogate.

The interface is a single guided flow rather than a dashboard. The result is a shortlist with reasons, not a score with no explanation.

Responsible AI and security

The matching runs on the device. That decision keeps the attendee’s answers out of a server-side profile and removes a data store that would have needed retention rules, access controls and a breach plan for a one-day event.

The tool does not claim to know who someone is, and it does not make an automated decision about them. It offers a ranked shortlist that the attendee can ignore. The reasons attached to each result are part of the design: a recommendation that cannot explain itself is not useful in a room where people are already sceptical of AI.

Limitations

This is a demonstration build, not a production event platform. It was designed around one room, one programme and one day. The matching model is small, the content set is fixed, and there is no attendee account, no persistence layer and no integration with a ticketing or CRM system.

The shortlist is only as good as the session descriptions and the attendee’s answers. A vague answer produces a vague result, and the tool does not pretend otherwise.

Source availability

The source is Not public. The live build is available at elevatech-compass.kriz2phorus.workers.dev. The case study describes the decisions and the constraints; it does not publish the private working repository.

Shipped

Deployed, and linked from here.

Finished applications with a public demo. The source stays private; the running product is what you can use and inspect.

  • Methane Minder

    A live full-stack methane plume dashboard built on Carbon Mapper's published data.

    Deployed Open app

Research threads

In audit, not in the shop window.

These are named directions, not demos. They stay here without repository or product links until the audit is done and the work is safe to show.

  • Risk-based MFA and SOC triage

    A research thread on using risk signals to prioritise authentication challenges and security-operations triage.

    Source audit completed 2026-11-18

    In audit
  • Agent assurance harness

    A research thread on testing whether an agent's effective authority stays inside its approved authority.

    Not yet audited

    In audit