Lab

Side Projects

Things built for curiosity, not a job description. Each one started as a problem I kept running into and couldn't find a good solution for.

01◈

Swarm

A multi-agent market sentiment simulator. Feed it a news headline and a swarm of AI trader personas — quant funds, macro tourists, retail degenerates — react independently, then debate each other and update their positions.

Key detailRound 1 is independent — every agent forms a view without knowing what anyone else thinks. Round 2 is the debate. The emergent consensus is what no single prompt could produce.

Result

Emergent market consensus from competing AI agents

Stack

GoClaude API

View on GitHub ↗Full write-up →

02◉

Njia

Intelligent running route generator for Nairobi, Kenya. Generates looped routes from any starting point and enriches them with real-time weather mud-risk, shade scoring, crowd-sourced safety alerts, and M-Pesa-gated premium plans.

Key detailSafety scoring is time-aware: a darkness alert filed at 21:00 carries full weight at night but zero weight at noon. Hazard expiry is per-category — flooding clears in 3 days, missing pavement in 180.

Result

Route intelligence built for Nairobi

Stack

TypeScriptExpressNext.jsPostgreSQLRedisGraphHopperMapbox GLDocker

View on GitHub ↗Full write-up →

03◎

Radar

An AI-powered arXiv digest for developers. Fetches today's research papers, extracts semantic dependency graphs, scores each paper on build potential, and runs a cross-paper synthesis to surface the most actionable findings first.

Key detail`radar build <N>` takes any paper from your last run and generates a full project brief: name, tagline, problem, stack, MVP features, task list, and time estimate.

Result

From raw arXiv feed to ranked build opportunities in one command

Stack

GoClaude API

View on GitHub ↗Full write-up →