M.J. / 01
AI & Systems Engineer — Nairobi, KE

I build the backend that makes AI systems actually run.

Solo founder and full-stack engineer with ~3 years applied AI/software experience. I design and ship end-to-end systems — from real-time detection pipelines to bonded-cellular hardware — and I'm comfortable reasoning about model behavior, prompts, and edge cases at the level a training pipeline needs.

Kazumee

production — live

Autonomous AI clipping tool for streamers. Detects hype moments from live chat and audio using an adaptive per-stream statistical baseline (z-scores/EWMA — not fixed thresholds), then triggers a local desktop agent to pull the real OBS replay buffer, transcode, and publish the clip to a web dashboard in seconds — fully autonomous, no manual editing.

FastAPIPostgreSQLReact / Vite WebSocketsGroq (Llama 3.3)faster-whisper PyInstaller agentRailway / Vercel
View repo →

Zumi Link Pro

hardware — spec complete

Compact bonded-cellular streaming hardware: a wallet-sized encoder unit with four simultaneous carrier modems and a wireless HDMI transmitter dongle, built to replace $10k+ broadcast rigs at a fraction of the cost. Reached full v5 specification — firmware, bill of materials, and build documentation — using SRT with forward error correction for stream resilience.

Embedded systemsRF / cellular bonding SRT + FECHardware BOMFirmware

Gold Node

deployed — private beta

Private AI backend-as-a-service platform built for enterprise clients: a self-hosted stack giving companies their own retrieval-augmented AI layer over internal data, with workflow automation and model routing built in.

FastAPIPostgres / pgvectorn8n OpenRouterNginxDocker

Ideal Africa Safaris

live — client site

Full production website for a pan-African tour operator, built and shipped end-to-end for a freelance client: dated-departure package browsing, a custom safari request flow with camp selection, an auto-advancing photo hero, a social-impact section, and a chat widget for FAQ handling. Live at idealafricasafaris.com.

Next.jsTypeScriptTailwind CSS Framer MotionVercel
View repo →

Algorithmic Trading Bots

live demo account

Automated forex and crypto strategy execution, built from scratch and running on a live MT5 demo account. Implements London Breakout and New York session strategies on XAUUSD with custom lot-sizing logic, plus a separate BTC 4H breakout bot against the Binance API.

PythonMT5 APIBinance REST API Strategy backtesting

I'm a self-taught software/AI developer on top of a formal Electronics and Computer Engineering degree — currently pursuing a Master's alongside my own ventures. I learn by shipping full systems end-to-end rather than by working through courses, which is why my projects span real-time backend infra, applied AI, and hardware.

I spend a lot of time in the gap between "the model gives a plausible answer" and "the system is actually correct under real conditions" — debugging routing logic, race conditions, and edge cases that only show up once real data hits a real pipeline. That's the same muscle AI training work draws on: reading an output carefully, knowing why it's right or wrong, and being precise about the correction.

  • Backend Python, FastAPI, PostgreSQL
  • AI / ML LLM APIs, RAG, prompt design
  • Infra Docker, Railway, Vercel
  • Frontend React, TypeScript (developing)
  • Hardware Embedded systems, RF
  • Experience ~3 yrs applied AI/software

Precision

Comfortable tracing a bug or a bad output back to its exact root cause rather than patching symptoms — the ops notes across my projects are full of "root cause found," not guesses.

Written clarity

I document architecture, decisions, and trade-offs as I go, so handoffs and reviews are fast and unambiguous — a habit that transfers directly to writing clear evaluation feedback.

Independent judgment

Solo founder across four concurrent projects — used to making scoped, defensible calls without waiting on a team, and to being direct about what's working and what isn't.