I bridge actuarial mathematics and production-grade software. Whether modelling income risk for Nairobi's gig workers or backtesting alphas for WorldQuant, the discipline is the same: find signal, eliminate noise, build something that holds under pressure.

Python · FastAPI · Gemini 2.5 Flash · Prophet · Streamlit

A parametric AI insurance engine for Kenyan gig workers. Ingests M-Pesa PDFs via Gemini Vision OCR, runs a Prophet time-series forecast calibrated to Kenyan public holidays and school-fee cycles, then auto-triggers compensation when income falls below 80% of μ (the actuarial dip threshold). No claims adjuster. No paperwork.

→ Core formula: payout = min(src_cap, (μ − current_income) × 0.70)

Python

Standalone actuarial engine extracted from IDCS — portable income-dip detection and compensation calculation library.

HTML · CSS · JavaScript

Online tech consultation platform for hardware and software problem resolution. Deployed and live on Vercel.

Python · HTML · CSS

Event planning platform built for millennials. FastAPI backend with Vercel deployment.

• IQC 2026 (WorldQuant BRAIN) — Engineering and backtesting alpha expressions for the International Quant Championship. Focus: momentum decay, cross-sectional normalisation, regime-aware signal construction.
• LegalTech Case Management System — Digitising physical case records for Nairobi law firms. Full-text search, matter-level indexing, FastAPI backend targeting low-bandwidth environments.
• Income Data Engineering Pipeline — Longitudinal data collection spanning 500+ subjects; schema governance, cleaning logic, and aggregation pipelines.

Long-exposure photography has sharpened the same instinct I rely on in time-series work: distinguishing real signal from ambient noise. Most nights in Nairobi produce more light pollution than stars — so you learn to work with what you have, expose longer, and stack frames. It's not a coincidence that the same patience applies to backtesting a weak alpha or waiting for a Prophet forecast to converge.