Applied AI · London
I make enterprise financial data ready for AI, and build the AI that uses it.
At Bloomberg I own the Data License API Platform, serving enterprise clients worldwide, and lead its applied AI: systems that answer product questions, investigate data exceptions and help resolve incidents. Before that, enterprise data platforms at J.P. Morgan, Fidelity and Schroders.
BloombergJ.P. MorganFidelitySchrodersHSBCBlackRockPwCCiti
US patent application, model-driven data pipelinesIIT KharagpurIIM CalcuttaCQF
Essays
On making AI agents dependable in businesses that have to answer for the result.
Series · 2 parts
Agent harness design
A series on the layer between the model and the world: the product concerns that decide whether an enterprise agent is reliable enough to ship.
- 01The cushions an agent loses in production
- 02The authority surface
When to use decision models
Which AI workflows are bounded decisions, how to turn a model's probabilities into thresholds from the cost of an error, and how to evaluate before automating.
The authority surface
Agent harness design, part II. Autonomy is not a dial. What an agent may do alone has to be decided action by action, on the cost of being wrong.
The record nobody agreed to keep
Why accountability for an agent's actions falls between the protocol, the schema and the regulation, and the record a firm will have to construct itself.
AI for Finance (Lab)
Built on my own time to test the hard parts of AI in regulated finance: grounding, refusal, and evaluation tied to the firm's own record of truth.
Working build
Tax Policy Navigator
Right behaviour on 48 of 50 UK tax questions, yet only 195 of 261 cited claims, 75 percent, were fully supported by the paragraph cited.
Working build
Rule Grounding
Every FCA Handbook citation an agent makes, checked by three tests that use no model. The check an audit needs cannot be built yet: the API serves only today's rulebook.
Working build
Governed Data Eval
Same model, same question: £2,512,500 on the raw systems, £1,512,500 on governed data. The gap is one superseded account.
AI for social good
Two free maths platforms for students, each built and run single-handedly with AI.
400 students use them, free.
Free platform · lemmamaths.co.uk
Lemma
Free preparation for the TMUA university admissions test: 47 interactive lessons, 2,000 problems with step-by-step solutions and 18 timed mocks. No card, ever.
Free platform · olympiadhq.com
OlympiadHQ
Olympiad problems from around the world, curated by tier and theme, with a problem of the day and video lessons. Free, as a public good.
AI for mathematics
Research on open problems, done with AI under my direction. I have settled a number of them, fourteen posed on MathOverflow and one from Erich Friedman's list of unsolved problems, where he credits my proof. Every paper is on GitHub with the code that checks it.
Almost-squares in almost-squares
Which k × (k+1) rectangles split into smaller distinct ones: exactly 4, 10, 12, 14, 15, 18 and every k ≥ 20. Friedman's conjecture.
Two triangulations, one outline
From nine points on, two triangulations can share nothing but the five outline edges, answering Rote.
Does the smallest copy fit?
For many random points in an equilateral triangle, the smallest enclosing copy fits inside with probability tending to 13/48, not 1/2.
