Rafe Tariq

I build autonomous agents, and research AI that does more with less.

About

Hey — I'm Rafe. I got into maths early: I'd finished my maths qualifications, GCSE included, by the time I was 12 — all of it at what was then the second-worst-ranked school in the country. Straight after, I got into game development on the WebGL stack — and into hardware — early. That set the tone: I've been chasing technical problems ever since.

I studied Computer Science and Artificial Intelligence (Bachelor's and Master's), under Dimitar Kazakov and Leandro Indrusiak. Along the way I did AI and weather-modelling research at Imperial College London — on the geospatial side, under Rossella Arcucci and Sibo Cheng — and I've contracted for a string of companies and funds, building quant tools and algorithmic trading suites with the latest AI models.

I started the British AI Collaboration, a UK platform for knowledge transfer between university AI societies. Then I joined Entrepreneur First's LD20 cohort to build autonomous agents, where I built LabBrain — first automating university wet and dry labs, then expanding into insurance and finance agents.

These days I'm doing research at the University of Manchester, running The Agentic Line — a product lab shipping vertical AI agents — and advising Unlocked Homes and a cybersecurity startup. I like problems where a model has to be both correct and cheap, and where the right interface turns a research result into something people actually use.

Projects

Things I've built and founded. Most of the agent products live under The Agentic Line.

Research

Selected papers. Interests: model compression & efficient transfer, computational biology, AI safety, and geospatial / weather modelling.

  1. Certified Vocabulary Compression: Side-Information Codebooks for Zero-Shot Transfer over High-Cardinality Sparse Categorical Transformers Rafe Tariq, Hongpeng Zhou · University of Manchester · 2026 (venue TBD)
  2. Machine Learning for Spatial Transcriptomics Analysis in Human Cancers: Tackling Limited-Data Challenges Review paper · authors & venue TBD · 2026
  3. Technical AI Safety and the Public Evidence Base: A Survey of Alignment, Robustness, Interpretability, Evaluation, and Control Survey · venue TBD · 2026

Writing

Casual notes and essays — on agents, ML, and whatever I'm building.

  1. The Baby: Where the Computer Was Born 2026 · Manchester ran the world's first stored program in 1948 — so why isn't it a tech city? On red tape, brain drain, dead factories, and where the city sits in the history of computing and AI
  2. Taste Is the New Pedigree 2026 · why aesthetic signal is replacing the credential — read through Taste Labs' $18.5M round, the AI cafes, and looksmaxxing
  3. Two Years, Then the Reckoning: Keir Starmer's Record in Office 2026 · a factual, fully-referenced ledger of what the Starmer government delivered — and the promises that defined its fall

Talks

Talks and meetups I've spoken at.

  1. Artificial Intelligence in Surgery ↗ BAUS North West Urology Meeting (#NWU25), Hilton Hotel, Liverpool · January 2025 · seminar panel
  2. 3 Ways to Build Your Own Financial Assistant with Generative AI ↗ BCS Software Practice Advancement (SPA), London · December 2023 · for SingularityNET

Otherwise

Away from the keyboard, I used to pursue semi-pro kickboxing — years of training, sparring, and the odd competitive bout. It taught me most of what I know about showing up, taking the hits, and staying calm under pressure.

Contact

Best way to reach me is email — rafetariqyork@gmail.com. I'm also on GitHub and X.