I'm a Nairobi-based AI/ML Engineer who designs and ships production LLM systems — multi-agent pipelines, retrieval-augmented generation, and the backend infrastructure that keeps them fast and reliable. I like taking AI products from architecture diagram to deployed API, with a strong bias toward measurable, real-world impact.
role: AI/ML Engineer · Backend Architect
focus: LLM agents · RAG · API design · applied ML for emerging markets
location: Nairobi, Kenya
currently: Building multi-agent advisory systems on LangGraph + Gemini
learning: Model evaluation frameworks, on-device inference for low-connectivity regionsLLMs & Agent Frameworks
Modeling & Data
Backend & APIs
Data & Caching Layer
Frontend & Mobile
DevOps & Tooling
|
Multi-Agent Agricultural Advisory Platform Bilingual (EN/Swahili) advisory system for smallholder farmers, built on a nine-node LangGraph pipeline with Gemini, FastAPI, Supabase auth, PostgreSQL, and Redis caching. React Native/Expo mobile client with M-Pesa billing.
|
Distributed ML Training Infrastructure Dockerized worker system for distributed model training using gRPC, MinIO, and a finite-state-machine execution model, supporting both PyTorch and TensorFlow runtimes.
|
|
Custom AI Tool & Agent Integrations Production LangChain tools connecting LLM agents to real systems — vision-based diagnostics, parameterized database lookups, and safety-checked recommendation logic.
|
Payments & Government Services Integration FastAPI services with M-Pesa Daraja STK Push integration for subscription billing and public-sector payment portals.
|
| Principle | How it shows up in my work |
|---|---|
| Ship reliable systems | Design for failure — retries, cache invalidation, and clean error boundaries, not just happy paths |
| Evaluate before scaling | Prototype fast, but validate agent behavior and model outputs before they hit production |
| Build for real constraints | Design with intermittent connectivity, cost limits, and local context in mind — not just ideal conditions |
| Document as I go | Clear READMEs, API docs, and architecture notes so systems outlive my involvement |
