I build reliable software across frontend, backend, mobile, and AI systems—from architecture and APIs to polished interfaces and production deployment.
Abuja, Nigeria · Available for opportunities
Production software across mobile, web, backend, and AI. Each case study separates what I owned from what I collaborated on.
All 9 case studies →Cross-platform community app on Google Play: feeds, moments, realtime chat, events, leaderboards, and an offline interest classifier.
MobileChats· 1 / 6
SeamHealth’s customer feedback platform: multi-step feedback flows, complaint handling, billing and an operator admin — 400+ of my commits across three Angular codebases.
MobileAnalytics· 1 / 7
Next.js web client for Whistler: 52 routes across feed, communities, moments, messages and settings, sharing the mobile app’s API, tokens and error handling.
MobileHome· 1 / 4
FastAPI service that fingerprints visitors, links probable identities across browsers and scores risk built to keep public voting fair, with tests and AWS deployment scaffolding.
The classifier maps a post to one of 30 interest categories without a network request. Training happens offline in Python; the app ships only the exported weights and about 350 lines of TypeScript inference code.
The training script reported 99.0% precision@1 on held-out data. Re-running it showed why: each hand-written sentence was augmented into three near-copies before a random 85/15 split, so 2,199 of the 2,238 validation lines had a twin in training.
Splitting by sentence first gives 60–64% precision@1 across three random splits, against a 3.3% chance baseline. The data is 4,962 unique hand-written sentences across 30 categories. No real user posts have been evaluated yet.
Grouped by discipline, with where each one was used in shipped work.
Whistler Web and the SeamHealth blog in Next.js; Whistler Admin and Clientshot Public in React 19; Clientshot in Angular; component libraries used across 3+ internal products.
The Device Intelligence service in Python and FastAPI; REST API design and integration at SeamHealth; a signed webhook function for Whistler Admin; the Socket.IO realtime client in Whistler.
Whistler: a React Native + Expo app on Google Play with push, deep links, social sign-in and over-the-air releases.
Two Whistler post classifiers: an on-device fastText model in the app, and sentence embeddings (all-MiniLM-L6-v2, ONNX Runtime Web) in the browser; LLM evaluation and RLHF at SeamHealth — prompt design, rubric-based review, data annotation and human feedback on model responses.
GitHub Actions CI, Docker images, Playwright e2e, Vitest, Jest and pytest suites, Sentry, Mixpanel, Firebase Analytics, k6 load tests.
The same loop on every project, whether it’s a form field or an ML feature.
Clarify requirements, constraints, users, data, and the business goal behind the request.
Choose boundaries, state ownership, API contracts, data structures and dependencies deliberately.
Write modular, typed code with abstractions that earn their place.
Check behaviour with unit, integration and regression tests — including realistic failure cases.
Automate builds and releases, and stage them through environments before production.
Use crash reporting, logs and analytics to see how the software behaves for real users.
Iterate on evidence. Measure before claiming, and say so when something hasn’t been measured.
Only tools I’ve used on shipped work are listed.
Typed code, with pre-commit hooks as the first gate.
TypeScript · Husky + lint-staged
Every change reviewed before merge.
GitHub · Bitbucket
Unit and integration tests; end-to-end tests run against a mock API.
Vitest · Jest · Playwright · pytest
Lint, types, tests, dependency audit and build gate every pull request.
GitHub Actions · Dependabot
Reproducible web, container and native builds.
Vite · Docker · Gradle · EAS
Staged release channels; JS-only fixes over the air.
Vercel · Cloudflare Workers · EAS Update · Google Play
Crashes, errors and product usage in production.
Sentry · Mixpanel · Firebase Analytics
Load-test risky paths and feed findings into the next change.
k6
Most frontend bugs I’ve fixed trace back to one of three things: unclear ownership of state, validation in the wrong place, or an abstraction that was shared too early. The architecture work is mostly about preventing those.
Realtime features are easy to demo and hard to keep correct. The work is in what happens when the connection isn’t there — which, on mobile, is often.
Writing the screens is the smaller part of mobile work. The larger part is integrations, releases and knowing what happened after a release went out.
The useful question is rarely “which model is best?” It is “what is the smallest, cheapest, most predictable thing that solves this task — and how will I know if it works?”
2025 — Present
Remote · Freelance
Whistler
Lead engineer on the Whistler mobile app, web client and admin dashboard; sole engineer on the web client.
2023 — Present
Abuja, Nigeria · Hybrid
SeamHealth Group
Frontend engineer on the Clientshot platform; lead engineer on Clientshot Public; first author of the Device Intelligence backend and the SeamHealth blog.
2022 — 2023
Abuja, Nigeria · Hybrid
Nigeria Civil Defence Corps HQ
Built and maintained internal web applications.
Interested in working together, discussing an engineering problem, or exploring an opportunity?
Francis.dev31@gmail.com
