Skip to content
EE

About

I take products from an empty repository to something real people use. Not a slice of the frontend, not a ticket queue — the whole thing: the data model, the services behind it, the interface, the emails, the deployment, and the 2 a.m. production bug.

Six years in, most of that has been JavaScript and TypeScript across React and Next.js, with Python — FastAPI, SQLAlchemy, pandas — doing the work behind it. At Biddy I built event-driven microservices talking over NATS.io, centralised authentication behind Keycloak, modelled the data in PostgreSQL through Prisma, and shipped the lot with Docker and Kubernetes. Functionality alone was never the bar: observability, security and staying up are part of the job, and I was the one on call when they weren't.

Plenty of it has been built alone. Biddy went from zero to more than 10,000 users with me owning every layer. A marketplace for a Lagos client shipped in phases across ten weeks. A tuition academy in Bradford went from a paper flyer to a live site in seven days. That range is deliberate — I'd rather understand a product completely than specialise in one corner of it.

Lately most of my attention goes to two things: two-sided marketplaces, where the hard problem is trust rather than code, and AI systems that run against real inputs instead of a demo script.

Download CVPDF · updated August 2026
Emmanuel Esekhaigbe

Warsaw, Poland — available for client work

What I take on

Marketplaces & platforms

Two-sided products where the real problem is trust, not code. Roles and permissions, moderation queues, reviews, search that survives a hundred thousand rows, and a revenue gate designed in from day one rather than bolted on later. Event-driven services and centralised auth when the thing is big enough to need them.

Biddy · Fixam

Sites that earn their keep

Static-rendered marketing sites that load instantly on a bad connection and actually rank. Real content architecture, structured data, and forms that deliver — verified against the served HTML, not just the browser.

TestPortal Academy · Vision Speaks · Tee & Tee

AI systems in production

Language models wired into things that run unattended against messy real input — live audio, market data, user documents. The interesting work is everything around the model: capture, validation, fallbacks, and cost.

AI Interviewer · Polymarket scanner

Background

Education

  • MSc, Management & Production Engineering

    Warsaw University of Technology

    Feb 2026 — present

  • BSc, Mechanical Engineering — First-Class Honours

    Landmark University, Nigeria

    2018 — 2022

Languages

  • EnglishFluent (C1+)
  • PolishBeginner (A1), actively learning
Get in touch →