Now — Testimony Adekoya

Building

FAUE — African Fashion Intelligence Platform

Active · Pre-build

FAUE is a fashion intelligence platform built for African fashion consumers — specifically the Nigerian professional who navigates Owambe dress codes, Aso-Ebi coordination, and fabric-to-tailor styling. The core loop: a Style DNA Quiz surfaces your taste profile, then the system recommends cross-brand outfits (RTW) and fabric-specific bespoke styling suggestions you can act on immediately — shop links or a WhatsApp share straight to your tailor.

Right now I'm in pre-build validation phase: finalizing the MVP feature set, designing the Style DNA Quiz and its conversion into a user embedding, and sourcing the initial self-curated catalog. The hardest problem to solve is cold-start personalization without wardrobe photo uploads — the Style DNA approach is the bet.

Pre-build Recommendation Engine African Fashion NLP + CV

Intellign — Decision Intelligence Platform

Production

At DataBacked Africa I'm leading the architecture of Intellign — operators describe allocation goals in natural language, and a genetic algorithm engine processes 500+ resources against 750+ targets under competing constraints in minutes. Our first major deployment assigned medical graduates to healthcare facilities. Currently iterating on the SSE-based real-time progress streaming and building human-in-the-loop review workflows.

FastAPI Genetic Algorithms SSE Decision Intelligence

AfriVTON-Bench — Research

Ongoing · 2026

With Shiloh Oni, I'm building a benchmark that evaluates how state-of-the-art virtual try-on models handle African textile patterns — high-frequency prints, non-Western garment silhouettes, pattern blurring, and motif misalignment. We're conducting a representation audit of VITON-HD, DressCode, and DeepFashion, then measuring failure modes using FID, LPIPS, SSIM, and frequency-domain texture analysis.

Computer Vision Generative Models Cultural Bias Benchmarking

Reading

Designing Machine Learning Systems

Chip Huyen — O'Reilly

Invaluable for thinking about the full lifecycle — not just model training, but data pipelines, feature stores, deployment, monitoring, and the organizational patterns that make ML actually work in production. Directly applicable to Intellign and what I'm building at DataBacked.

The Alignment Problem

Brian Christian — Norton

Reading it through the lens of cultural representation — if alignment is fundamentally about encoding human values, whose values are being encoded? How do you build reward functions for fashion in Nigerian social contexts that Western training data can't capture?

The Mom Test

Rob Fitzpatrick — Independently published

Rereading this before FAUE user interviews. The whole book is about how to have conversations that give you real signal instead of polite lies. Critical for validating whether the Style DNA → Recommendation loop actually drives action before building it.

Thinking About

Cultural context as a first-class signal in recommendation systems

Most recommendation engines optimize for engagement on globally-uniform behavioral patterns. But Nigerian fashion is context-dense — the same person dresses completely differently for corporate environments, Owambe parties, and Aso-Ebi group events. How do you model that context-switching as a feature? What does a "situation-aware" style embedding look like?

MLOps maturity gaps in African AI startups

After working at DataBacked Africa and Obscura Finance, I'm noticing a consistent pattern: African AI companies can build impressive models but the monitoring, retraining pipelines, and observability stacks are often an afterthought. The gap between "model shipped" and "model maintained" is where most production ML fails. Writing something about this.

What does "African AI" actually mean at the system level?

Is it just models trained on African data? Or is it systems designed with African infrastructure constraints — unreliable internet, lower-spec mobile hardware, different social trust structures? AfriVTON-Bench is partly about making this concrete: cultural robustness as a measurable property, not a vague aspiration.

The right scope for an ML engineer vs. a researcher

I exist at the intersection of applied engineering and research. The engineering side demands shipping working systems; the research side demands careful null-hypothesis thinking. The MMIBC work and AfriVTON-Bench forced me to develop both muscles simultaneously. Wondering how long that tension is sustainable, and whether the right long-term path is to lean one direction.