Building scalable solutions that solve real problems with frontier technology. 6+ years experience serving tens of thousands of users.
I started building frontends in 2019 and haven't stopped shipping since. Over 6+ years I've gone from React and React Native apps to full-stack systems with FastAPI, Django, and PostgreSQL — and more recently into AI/ML, fine-tuning models and building agent workflows.
Right now I'm CTO at Navismart AI, building an immigration platform that uses AI agents to guide users through complex application processes. I'm also building Galexia, an educational platform that turns questions into interactive visualizations — flowcharts, 3D molecules, animated canvases, 13 types in total.
Most of my work has been for East African businesses and users. I've helped SMEs go digital for the first time and built consumer apps used by 20k+ people on networks where every kilobyte matters. That shapes how I think about performance, UX, and what "good enough" actually means.
I use AI tooling daily — Claude Code, Google Antigravity, GitHub Copilot — and I think seriously about where these tools help and where they don't. I also fine-tuned a Gemma 3 model for Kikuyu-English translation, a language spoken by 8M+ people with almost no AI tooling. I build what's missing.
Leading a team to build navismartai.com, an AI-powered immigration platform for individuals and agencies.

Built three apps in the Jobzy ecosystem (Talanta AI App, Service Provider App, Client App).

Founded a software consultancy delivering custom solutions for web and mobile apps.
Conducted a 2-week consulting contract reviewing code and UI/UX for M'Network and M'Butic platforms.

Built award-winning multivendor e-commerce platform for construction materials (1st place Kenya E-commerce Award).

Built three apps in the Jobzy ecosystem (Talanta AI App, Service Provider App, Client App).

Built a service procurement platform end-to-end, covering both web and mobile.

Built a laundry service app using React Native and Expo.
Contributed to Zumaridi, a video conferencing platform.
Implemented solutions using React, React Native, HTML, CSS, and SCSS, applying modern frontend practices.
Conducted independent research and development in AI and machine learning.

AI-powered immigration platform serving 2k+ users across 50+ countries. Features eligibility checks, smart form filling, AI guidance, multilingual support, and secure application tracking.

AI-powered educational platform that transforms questions into interactive visualizations. 13 visualization types including flowcharts, timelines, concept maps, 3D molecules, and animated canvases. Uses AI to dynamically generate the right visualization for any topic.

A fine-tuned Gemma 3 4B translation model for bidirectional English-Kikuyu translation. Trained through 4 progressive stages over 55+ hours on 141k translation pairs including Bible, FLORES-200, Bloom Library, agricultural terminology, and synthetic data. Features LoRA fine-tuning achieving 0.746 final loss.

An AI Powered Job Search React Native mobile app for iOS, Android, and Huawei devices, enabling seamless job searches with advanced filters, personalized job alerts, and intuitive navigation. You can use AI to generate Resumes, CVs, and Cover Letters as well as job application emails.

Custom farm management software with mobile application and admin site. Helps clients manage their farm activities, track expenses, income, and overall farm operations.

Construction e-commerce platform connecting clients with service providers. Platform includes vendor dashboard, client interface, and mobile applications.

Mobile application for purchasing construction materials and requesting services from providers. Integrated with the main M-jenzi platform.

Comprehensive vendor management dashboard where vendors can register and manage their products. Complete e-commerce management solution.


A concurrent task scheduling and execution system built with Elixir/OTP patterns. Features GenServer-based task management, ETS-backed high-performance storage, DynamicSupervisor for worker processes, and fault-tolerant supervision trees with automatic retries.

A real-time task management API built with Phoenix Framework. Features RESTful JSON API, Phoenix Channels for WebSocket-based live updates, Ecto changesets for validation, and a complete task lifecycle management system.

An open source feature rich, developer friendly, modifiable chatbot UI and backend for Claude AI. Supports projects.

Implementation of OpenAI's open source Whisper model for audio transcription using PyTorch.

Implementation of Tortoise TTS for high-quality text-to-speech conversion.

Implementation of Generative Adversarial Network for fashion image generation.
A working developer's honest assessment of where AI-assisted development actually stands — what works, what doesn't, and what most people get wrong.
How I trained a bidirectional English-Kikuyu translation model using LoRA, progressive fine-tuning, and 141K parallel translations.