Tool-Using Agents
Agents built as production software: typed tools, fail-closed guardrails, and a named operator behind every change they make.
Open to Principal, Staff & leadership roles Nairobi, Kenya (Remote)
Full-stack AI engineer / Backend Systems & Production AI. Leads backend and AI engineering on a streaming platform's Java 21 and Spring Boot services: payments that never double-charge, auditable revenue sharing and real-time analytics. Builds AI the same way, owned from design to support: tool-using agents that make real, auditable changes, and small, quantized models that meet tight latency budgets on CPU.
01 Core Engineering Focus
I build backend systems and take AI into production, and own each system from design to support: payments that never double-charge, agents that act within guardrails, and models sized to the latency budget and hardware they run on.
Agents built as production software: typed tools, fail-closed guardrails, and a named operator behind every change they make.
Trains and distils models on GPU, then serves small, quantized models on CPU that meet tight latency budgets without wasted compute.
Java and Spring Boot services that charge exactly once, split revenue with a full audit trail and stream events into analytics, owned from API design to production support.
02 Tech Stack
The languages, data systems, and ML runtimes behind production AI.
03 Experience
2018 — 2024
Upwork · Remote, international clients
2023 — 2025
Zone01 (Berrijam AI projects)
2023 — 2025
Zone01 · Kisumu, Kenya
2025 — Apr 2026
BBR (Big Bold Red Agency) · Nairobi, Kenya
Apr 2026 — Present
BBR (Big Bold Red Agency) · Nairobi, Kenya
— Present
BBR (Big Bold Red Agency) Nairobi, Kenya
—
BBR (Big Bold Red Agency) Nairobi, Kenya
—
Zone01 (Berrijam AI projects)
—
Zone01 Kisumu, Kenya
—
Upwork Remote, international clients
Years at 0 went to client work in private, confidential Gitea and Bitbucket repositories, which GitHub does not count.
04 Featured Projects
Safaricom's video and music streaming platform on web, Android, iOS and TV. Built its revenue-sharing and analytics services, designed the payments rail for M-Pesa and carrier billing, led the upgrade to Java 21 and Spring Boot 4, and delivered features on Baze Web.
Payments that never double-charge, and revenue shared with a full audit trail
The entertainment hub powered by Safaricom. Built content moderation with BBRaiN AI for its operations team, integrated Safaricom single sign-on into the CMS, and configured BBRaiN as a separate tenant with its own catalogue and rules.
BBRaiN moderates its content and powers its search and recommendations
One AI engine for search, recommendations and personalization across all media: a Rust inference engine, a GPU training factory and a signed model hub that delivers models into each client's own cloud (VPC), run through an MLOps loop from training to drift monitoring. Replaced Recombee and Algolia on Baze (Safaricom platform).
About 200,000 lines of Rust, backed by about 4,000 tests
AI sentiment intelligence for African markets. Reads brand sentiment in Swahili, Sheng, Pidgin and Amharic languages.
Turns conversations into a brand-health score, trends and benchmarks
05 Workflows
How the systems I build move requests, money, data and models: each one an engraved diagram you step through, every hop captioned with what happens there and what can go wrong.
14 parts · 11 steps
Large teacher models are distilled into one small student, trained for INT8 and served on CPU, shipped as signed bundles into each tenant's own cloud, and watched in production until drift starts the loop again.
24 parts · 16 steps
An order on a Kubernetes platform is checked at the edge, recorded once with its event, carried by Kafka through services that reserve the stock and authorise the card, with the payment captured only when the order ships, and traced from checkout to dispatch.
32 parts · 15 steps
Sign-in, the home screen, a checkout and a play request from the web, phone and TV apps, each through its own backend for frontend, with Kafka carrying the events, change data, heartbeats, activity and logs behind them.
12 parts · 10 steps
A model is built, checked, released in stages and watched in production. When the live data drifts away from what it was trained on, an alert starts the loop again.
06 Contact
Open to Principal, Staff and engineering leadership roles, and to teams that need backends and AI running in production, on their own hardware and in their own cloud.