Senior backend engineer, 11+ years on distributed systems: Kafka, PostgreSQL at scale, financial-grade reliability. I learned it where a mistake costs money. Lately I've brought that same discipline to applied AI, RAG and multi-agent systems.
11+ years of professional IT experience. Open to the right senior role, remote or relocation.
Scroll for the details, or jump straight to the resume.
Three pieces of work that show what I do: keeping a payments queue correct, and keeping a 30M-row database fast as it grew.
On a payments platform, a message that gets lost or processed twice is real money moving the wrong way. I was responsible for making that reliable under load: catch the bad messages, retry them sensibly, and make sure replaying one can never double-charge anyone. The number I watched most closely was the backlog per partition, because one bad message left unhandled can quietly hold up everything behind it.
I scaled a PostgreSQL database past 30M rows and kept it quick: splitting the data up by region, indexing it for the queries people actually ran, and sending reports to read-only copies so heavy reporting never slowed down live work. The slowest dashboards dropped from multi-second to near-instant. A separate service held 12K+ requests per second in testing.
I designed and built a service that turned 15M+ company records into one clean, consistent set of countries, states, and cities. People type the same place a dozen different ways; this made them all line up so every search and report downstream could trust the data. Earlier in my career, I built regulated systems from scratch where the input could never be taken on faith.
The same discipline carries straight into AI. FinResearch AI is a multi-agent system where every step emits structured JSON, so the agents answer from what they retrieved instead of guessing. When the evidence is thin, it says so rather than invent an answer.
Open FinResearch AI on Hugging Face →On a payments system, a bad message can't be dropped or processed twice. I built the workflow to catch it, set it aside, retry it on a backoff, and replay it once. The diagram below shows that flow.
The message is processed exactly once, nothing lost or duplicated. That's a dead-letter queue with bounded retry and idempotent replay, the pattern I used on a payments platform.
A couple of things about how I work, and what I build when it's just me.
I read every message myself and I reply. Open to the right senior role, remote or relocation.
Everything as plain text. Senior Full-Stack Engineer, backend-heavy, 11+ years of professional IT experience, distributed systems with a real, growing practice in applied AI.
Download my resume (PDF) ↓An event-driven race-strategy service in the language I trust under load: a reactive API publishes a request to Kafka, a consumer runs a deterministic simulation, and the result is written back exactly once. The public, NDA-free analogue of the financial-grade reliability work below: bounded retry with backoff and idempotent processing, so a bad message is retried sensibly and never processed twice.
read the code →An agentic RAG assistant for Formula 1 questions, built with production hygiene: auth, rate-limiting, tests, structured logging, caching, graceful degradation, CI/CD, and Pydantic-typed state. A single agent routes between a vector store and live web search, then answers over what it retrieves. It started because I follow the sport, and it became my sandbox for getting the engineering around an agent right.
open the live demo →A LangGraph multi-agent system I architected and deployed: a manager fans out to specialized research agents in parallel, then an analyst and a reporter write a verdict over what was retrieved. Every step emits structured JSON outputs so the agents stay grounded instead of making things up. Built as my submission to the SuperDataScience CP044 community project.
open the live demo →A reference architecture for watching an AI agent's tool calls: it observes calls, runs an advisory policy gate and a tamper-evident audit, all evaluated out-of-band, never in the agent's hot path. It observes and records; it never decides what runs. Written in the kind of Java I trust to hold up under load.
An early build to help with the tax and visa side of moving between countries, a problem I am living through myself right now. Still pre-release.
Roles by public label, no client names. 11+ years across financial, banking, and data-intelligence systems.
Contributing to a pilot that replaces spreadsheet-heavy plant work with AI you can trust. It answers from the actual procedure documents, and when it is not confident enough it says so and cites the source instead of guessing.
Designed and built a data-cleanup service across 15M+ records and scaled the database past 30M rows so it stayed fast under heavy reporting. The slowest dashboards dropped from a wait to near-instant, 60%+ faster.
Built the payment workflows where correctness mattered most, so a bad message is caught, retried, and replayed cleanly without ever double-processing. Also built a service that held 12K+ requests per second in testing.
Built the secure login and anti-phishing layer in Java under real regulatory and performance constraints. New from scratch, and it had to be right the first time.
The same three accounts everywhere, cross-linked, so a reviewer who clicks finds exactly what the copy promised.
The code behind the claims, public and in the open.
The full work history, employers public by name.
connect on LinkedIn →FinResearch AI, a running multi-agent Space.
open the live demo →A RAG assistant for F1 questions, live.
open the live demo →Open to senior roles, remote or relocation · 11+ years of professional IT experience