§ /story · log entries
A CV lists roles. This page is the bits that did not fit on it — five entries, newest first, written so a stranger can pick up cold.
AI Engine Summer Hack 2025. I worked on backend and LLM orchestration for Moments AI — a personalised content generation platform. Team of four, 48 hours, live by Sunday.
Hackathons are a forcing function for shipping. Two days with a fixed team and no time to over-engineer is the opposite of how production work feels, and it is the fastest way to find out where my own bottlenecks actually are.
viberole.app shipped. Auth, generation, Stripe checkout, post-purchase delivery. The LLM layer is provider-agnostic — I can swap or A/B between OpenAI and Anthropic without touching product code, and there are prompt and cost guardrails so a single user cannot blow up the bill.
I wanted to see how far a solo engineer can take an AI product in 2025. The answer turns out to be "pretty far," once the LLM plumbing is abstracted; most of the work after that looks like normal product engineering.
I built an automated emissions calculation system in Python. It reads unstructured documents — invoices, supplier reports, utility statements — and returns a structured emissions breakdown, with citations back to the source document.
The LLM extracts and classifies; deterministic code does the arithmetic. Keeping those jobs separate is most of what made the system reliable, and is now how I think about LLMs in any pipeline.
Software Engineer at Kiya.ai. Java and Spring Boot REST APIs for AML and core banking — regulated, real money, real on-call. The work that taught me the most was not greenfield. It was bad SQL, broken integrations, and reading the system carefully before changing it.
I improved API response times by ~20% through service-level optimisation, and SQL query performance by ~15% through indexing, query review, and Hibernate fixes. Those numbers came from many small changes; none of them felt heroic in the moment.
B.E. Computer Engineering, First Class. Starting from zero never bothered me much — I break complex problems into smaller parts, ask questions early, and build understanding piece by piece until the system makes sense.
It is how I picked up AI during my MSc, how I approach unfamiliar codebases at work, and how I will approach the next role.