Distributed Systems · Java · Elixir · Agentic AI
I build back-end systems that hold up under real load, and I bring that same engineering discipline to AI. Open to relocation across the EU and the UAE.
I have spent 11+ years building back-end systems that hold up under real load, and the last stretch bringing AI into that same engineering discipline. The backbone is distributed systems: event-driven services on Kafka, 30M+ row PostgreSQL, idempotent recovery, and the service contracts that keep things honest when traffic spikes, learned first on financial and banking platforms where a mistake costs money. More recently I have been working on AI-assisted applications at Agilent and building multi-agent and retrieval systems in my own projects.
Contributing to a manufacturing-AI effort that is replacing Excel-heavy plant workflows with grounded applications. Prototyping a retrieval-augmented layer over standard-operating-procedure documents, with schema validation and guardrails so the output is auditable on a factory floor, not just plausible.
Designed and owned a geo-standardization service that normalized 15M+ company records into canonical shapes, the framework every downstream workflow runs through. Scaled PostgreSQL past 30M rows with partitioning and read replicas, and cut dashboard latency by more than 60%.
Owned the Kafka workflows where correctness mattered most: a dead-letter strategy, bounded retry with backoff, and idempotent replay, so operators could recover from a bad message without double-processing a financial transaction. Built an OpenResty layer holding 12K+ TPS in test.
Built the secure login and anti-phishing layer for a Kenyan bank's first online-banking platform, in Java under real regulatory and latency constraints. It was greenfield, and it had to be right the first time.
The interesting work was never the part that demoed well. It was the part that had to be right the first time, and stay right under load.
Two are live and answering to real users. Two are in active development, where the interesting engineering is still happening.
A multi-agent market-intelligence system that researches a company end to end, then hands back a report you can trust. Every step is Pydantic-validated and guardrailed, so the agents stay on a schema instead of free-styling, the same discipline I bring to back-end work.
open on hugging face →An agentic demo that answers Formula 1 questions by reasoning over live data, my sandbox for tool-use and orchestration patterns before they go anywhere serious.
try it →Agentic security middleware that sits in front of LLM traffic and checks it against the OWASP agentic-security risks, in the financial-grade Java I trust under load.
in development →Cross-border tax and visa compliance for people who move countries, the problem I am living myself. Built on Elixir and Phoenix with Oban Pro for the durable, idempotent jobs, and LangGraph for the reasoning over messy, jurisdiction-specific rules.
in development →Eleven years has narrowed this down to the things I actually trust in production. Grouped by what they do.
Ask about the systems I have built, my AI work, or how a move to your team would actually work. It answers first-person, plainly, and in my own words.
Ask me anything about my work.
I will keep it straight, in my own words. Pick a thread below or type your own.