I'm a Full-Stack Engineer with 4+ years shipping production systems in React, TypeScript, and Node.js. I don't compete on "I know React" — I compete on building systems that scale, cost less, and reduce complexity, backed by quantified impact.
- Systems over features. I approach architecture through domain-driven design — mapping entities, invariants, and state machines before writing application code.
- Invisible performance. Type-safe abstractions, sub-300ms real-time latency over 3G, and infrastructure cost cuts users never notice because it just works.
- AI-augmented building. I build developer tooling that makes LLM workflows cheaper and faster — local-first, BYOK, zero hosted infrastructure.
- Async-first, open-source ethos. Nigerian connectivity made me master clear written communication and zero-cost user economics.
Open To
Remote Full-Stack / Front-End roles•Measured pivot toward Cloud / Backend / DevOps•Low-friction collaboration (WAT / EU / Africa overlap)
I build with AI rather than train models — shipping local-first, BYOK tooling that makes LLM workflows cheaper, faster, and privacy-preserving.
| Domain | Proficiency | Details |
|---|---|---|
| LLM Context Optimization | Advanced |
Built tomd, a desktop GUI converting documents to token-efficient Markdown before they hit an LLM's context window |
| BYOK / Client-Side AI Integration | Advanced |
Architecting Sift — a read-only AI research sidebar with bring-your-own-key Gemini integration, no hosted backend |
| AI-Assisted Development | Advanced |
Daily-driver workflows for spec-writing, refactoring, and complexity reduction across the stack |
| Prompt & Schema Design | Intermediate |
Zod-validated structured outputs, deterministic query compilation (Scholarle), edge-case-aware AI flows |
| Privacy-First Architecture | Advanced |
100% on-device processing, zero telemetry, MIT-licensed tooling by default |
🗂️ tomd — Drag & drop anything. Get Markdown.
A tiny, open-source desktop app that turns PDFs, Word docs, spreadsheets, slides, and images into clean, token-efficient Markdown — the format LLMs natively speak. A transparent GUI over Microsoft's MarkItDown CLI: it finds or installs the binary in a private venv, shells out file-by-file, and saves the .md next to the original. No server, no accounts, no uploads.
| Stack | Desktop GUI · Microsoft MarkItDown · Python venv orchestration · Subprocess pipeline |
| Scale | 13+ input formats (PDF, Office, images w/ OCR, audio, HTML, archives, EPub, Outlook) |
| Performance | Sequential queue, live per-file progress, ~2–3s/document conversions |
| Security | 100% on-device, zero telemetry, private venv (global Python never touched) |
| Impact | Cuts LLM context overhead by converting documents before they reach the chat box |
| Repository | github.com/thekiwidev/tomd |
Built to solve a real pain: attaching a .docx or .pptx to an LLM burns context on formatting overhead before the model reads a single idea. tomd does the conversion locally and inspectably — every shell command streams to the screen, the binary in use is shown in the footer, and one folder delete undoes the entire setup.
🔍 Sift — Read-only AI research sidebar (in development)
An open-source Chrome extension that turns any page into a research surface — a read-only AI sidebar with no browser automation and a strictly client-side architecture. BYOK model with Gemini as the primary provider, so users pay zero hosted-infrastructure cost.
| Stack | TypeScript (strict) · Manifest V3 · React 19 · Zustand · Zod · Tailwind · Vite + CRXJS · Bun |
| Scale | Multi-scope context (page / selection / tabs) with token-budget management |
| Performance | Client-side-only, lazy context assembly, no background polling |
| Security | Git-ignored local key storage, CI secret scanning, no data exfiltration |
| Impact | Zero-cost research assistant — open-source, MIT-licensed, no backend to run |
| Repository | github.com/thekiwidev |
Designed against a full technical spec (architecture, domain models, edge cases, token-budget math, privacy posture) before a line of app code — domain-first, as I build everything.
📦 Momentum Logistics Service (MLS) — Multi-carrier logistics platform
A full-stack logistics platform for a Poland-based client, integrating multiple carriers and payment gateways with KSeF-compliant invoicing.
| Stack | Next.js · Bun · TypeScript · Node.js · PostgreSQL · Prisma · Tailwind |
| Scale | Multi-carrier (FedEx + DHL eCommerce APIs), dual payment gateways |
| Performance | SSR + Edge caching, Zod runtime validation, 15s gateway timeout |
| Security | Webhook signature verification, polling fallbacks for transactional availability |
| Impact | Drove a 40% increase in contract value from enterprise clients |
| Repository | Portfolio case study |
Dual-gateway payment engine (Stripe + PayU Poland) with verification and fallbacks ensures transactions stay available even when a provider degrades.
🎓 Scholarle — Google Scholar query builder
A smart query builder that compiles visual boolean filters into valid Google Scholar search syntax, handling the special cases most tools silently break on.
| Stack | Next.js 15 · React 19 · Radix UI · Tailwind |
| Scale | Boolean grouping, journal filtering, quote handling, length-limit guards |
| Performance | Deterministic query compilation, client-side |
| Security | No data collection — pure client-side URL generation |
| Impact | Prevents silent failures when queries exceed Google's length limits |
| Repository | Portfolio case study |
Solves the silent-failure problem: when queries exceed Google's limits or grouping is malformed, most tools fail invisibly. Scholarle compiles deterministically with guards.
Jul 2023 – Dec 2025 · Remote
Owned the front-end of a legal-tech platform spanning 14 visa categories and real-time legal interviews.
- Built a dynamic validation engine in TypeScript + JSON-schema handling conditional logic across 14 visa categories — cut 2K+ lines of redundant code and dropped form submission error rates 45%.
- Built FiBi's front-end on a WebSocket voice architecture with custom jitter-buffer logic, holding sub-300ms latency over 3G for live legal interviews.
- Added client-side audio compression, cutting server egress costs 30% with no data loss.
- Built a Tailwind + Headless UI design system (atomic design) that cut design-to-production handoff 70% at 100% WCAG 2.1 AA.
TypeScript React WebSockets JSON-Schema Tailwind Accessibility
Oct 2022 – Present · Remote
Co-pioneered an e-commerce platform connecting US diaspora communities to home-country products — consumer app, driver delivery system, and marketing site.
- Architected offline-first React Native consumer + driver apps with Zustand + AsyncStorage, lifting retention +22% in low-connectivity regions.
- Built a TanStack Query data layer with cache invalidation and request dedup via Axios interceptors, cutting redundant API calls 40%.
- Drove monthly infrastructure cost from $2,000 → $600 — $16,800/yr saved — while keeping real-time order tracking performant.
React Native Zustand TanStack Query Firebase Offline-First Cost Optimization
| Recognition | Details |
|---|---|
| 💰 Infrastructure Cost Cut | Reduced monthly infra spend ~70% ($2K → $600), saving $16,800/year |
| 🧩 Complexity Reduction | ~60% complexity + 45% error-rate reduction on a visa-form validation engine |
| ⚡ Real-Time Performance | Sustained sub-300ms WebSocket latency over 3G for live voice |
| ♿ Accessibility | 100% WCAG 2.1 AA design system; 100-Lighthouse portfolio |
| 📦 Open Source | Shipped tomd — local-first, MIT-licensed document→Markdown converter |
University of the People DEAC-accredited
Product School
learning:
- Cloud / Backend / DevOps (measured pivot)
- Systems design & algorithm foundations (B.Sc. CS)
building:
- tomd # local-first document → Markdown converter
- Sift # BYOK AI research sidebar, client-side only
exploring:
- Token-efficient LLM workflows
- Zero-cost, no-hosted-infra developer tooling
open_to:
- Remote Full-Stack / Front-End roles
- Cloud / Backend / DevOps opportunities
- Low-friction async collaboration (WAT · UTC+1)


