researching and building ai systems
Principal AI Engineer at Deliverance AI — taking AI into production.
Previously founding engineer at Nur Opus, an AI alignment research lab.
PhD Biomedical Engineering, MEng Mechanical Engineering.
stax.strath.ac.uk — PhD Thesis →I build the measurement layer for AI — calibrated LLM-as-judge systems, evaluation pipelines, and benchmarks that turn "is this any good?" into something rigorous and repeatable.
My research digs into how language models actually behave — where they reason well, where they fall down, and how to make them more reliable and aligned. Recent work includes a paper on AI sycophancy, currently under review.
Academic research for everyone — drop in a question, agents decompose it, search papers, and build a knowledge graph in real-time. Export the citations and summary, and share knowledge trees with friends.
arborinquiries.com →Read the latest in AI research — track papers across labs and topics with daily automated ingestion and curation. The perfect size for a morning read.
layerthelatestinalattice.com →Multiplayer prompt engineering adventure — co-op players defeat AI bosses through freeform prompting as the LLM boss degrades. Every single run is entirely original through GenAI.
coming soon →What do you really wish you did more? — break down any aim into a plan and go on a streak to achieve it.
coming soon →Kid-friendly classroom Q&A — kids ask questions, get simple safe answers powered by AI.
askaroo.co.uk →Ad hoc work with Ben Clarke at A2A Net — building a modular agent ecosystem using Google's Agent2Agent protocol. Re-architected a LangGraph search agent into OpenAI's Agents SDK to prove cross-framework interoperability. Containerised with Docker, deployed to Google Cloud Run.
github.com/FlynnLachendro/TavilyAgentOpenAI →Most of my work lives in private repos — the projects above are apps I'm actively working on. If you would like to see one of the not-yet-deployed apps you can see here then please reach out and I'll show you what I've got so far. Anything you see on here is in MVP stage or better, just not quite fully polished and shipped.