Three projects, each chosen to show how I think about frontend systems — architecture, composition, and the discipline of shipping production interfaces.
Case 0101 / 03
TEDxAdoEkiti
Software Engineer · 2024 / 2025
Case 01
TEDxAdoEkiti
TEDxAdoEkiti
Overview
The official web presence for TEDxAdoEkiti — an editorial, event-grade site built to carry the identity of the conference across speakers, program, and ticketing.
Context
TEDx events live and die by first impression. The brief was a site that felt like a designed publication: quiet typography, a clean program flow, and content that scales as new speakers and sessions are announced.
Problem
Static event pages age quickly and rarely survive a re-brand or a new lineup. The site had to behave like a small CMS-backed product — fast to update, rigorous about layout, and durable across editions.
Role
Software Engineer · 2024 / 2025
Engineering Breakdown
01Editorial layout system with a strict type scale for headlines, speakers, and program blocks.
02Component architecture organized around Speaker, Session, and Sponsor primitives that compose into any page.
03Content pipeline structured so new speakers and program items ship as data, not as new pages.
04Performance: image policies, route-level splitting, and prerendered pages for share-heavy surfaces.
05SEO and social sharing tuned per page — every speaker and session has its own metadata and preview card.
06Accessibility pass on color contrast, focus order, and keyboard navigation across the program flow.
Tech Stack
Next.jsTypeScriptTailwind CSSFramer Motion
Outcome
A publication-grade event site that reads as intentional, loads fast on low-bandwidth networks, and can carry future editions without a rebuild.
A high-precision operations console for the Ayede 330kV / 132kV substation of the Transmission Company of Nigeria — real-time parameter tracking, grid safety alarms, and an integrated operator advisor.
Context
Substation operators log hourly readings across feeders, transformers, and area loads. Existing tools were paper-first and error-prone; the console had to be Apple / Forbes-minimal but production-grade under 24/7 use.
Problem
Design a single operator interface that captures readings without friction, locks down inputs when equipment goes offline, and surfaces critical alarms — while remaining calm enough to sit on a control-room screen for an entire shift.
Role
Engineering Intern · 2026
Engineering Breakdown
01Operator entry form with progressive validation, temperature alerts, and out-of-range indicators.
02Auto-prefill and line safety locks — offline equipment forces inputs to a warning-red state and zeroes measurements.
03Real-time polling dashboard: 10-second background loops keep charts, stats, and alarms fresh with no reloads.
04Active safety alarms scanning for 300–360 kV boundaries, ≥ 80°C transformer windings, and missing shift logs.
05Three-page landscape PDF operational packs plus Excel exports partitioned by voltage grid and battery.
06Integrated operator advisor (Favour Alfred) grounded in live database state via Gemini 2.5 Flash.
A production-grade control-room console that replaced brittle paper workflows with a calm, auditable interface — Supabase-backed, RLS-secured, and synced to the team's Google Sheet of record.
Case 03
Milo
Milo
Overview
An AI language companion that makes you fluent through conversation, not memorisation — a persistent, memory-aware chat and voice partner powered by Gemini and pgvector.
Context
Most language apps drill grammar. Milo was built around a different thesis: real fluency comes from being spoken to like a person, with immersion that scales to your level and corrections that never break the conversation.
Problem
Design an AI companion that actually remembers you across sessions, adjusts language immersion to proficiency, and corrects grammar in the background without interrupting the flow of dialogue.
Role
Founder & Engineer · 2025 / 2026
Engineering Breakdown
01Persistent cognitive memory — facts, preferences, and emotional context extracted per turn, embedded, and stored in pgvector.
A working AI companion product with persistent memory, adaptive immersion, and a voice mode — architected so long-running conversations stay coherent without ballooning context windows.
