Back to the homepageThe work 
Featured caseCredit changes tested in 1 to 2 hours, down from five working daysA QA agent that reads a lender's live credit decisioning setup, finds where it has drifted from the requirements, and generates and runs the test cases before a change goes live.1 to 2 hoursChange request to tested, from 5 to 6 days16 to 4Manual checking, hours a week per analyst3Environments checked per runUnder 1 hrTo reconcile requirements with the live setup, down from about 3 daysCredit decisioningRead the full case
What I have built, case by case.Every figure says where it came from.
A few figures are confirmed by the client. Most are my own estimate from the work, and each one says which. References under NDA, on request.
Featured caseCredit changes tested in 1 to 2 hours, down from five working daysA QA agent that reads a lender's live credit decisioning setup, finds where it has drifted from the requirements, and generates and runs the test cases before a change goes live.1 to 2 hoursChange request to tested, from 5 to 6 days16 to 4Manual checking, hours a week per analyst3Environments checked per runUnder 1 hrTo reconcile requirements with the live setup, down from about 3 daysCredit decisioningRead the full caseCredit and lendingLong reads, four to five minutes each.
- HundredsTest cases per change, in minutesAn AI-enabled system that reads a lender's credit requirements, builds synthetic applicants for every path through the engine, and checks the engine's decisions before a change goes live.Credit decisioning
- 6 to 2Hours per client to pull together the policy rules and requirementsA system that connects to the places where credit requirements actually live, and, guided by the analyst, pre-fills one standard requirements document for every credit decisioning project.Credit decisioning
- 2 weeksTo a running product, from monthsA complete, ready-made personal loan product for fintechs and new lenders entering a new market: the credit logic, the scorecards and the requirements, already configured in the decisioning engine.Credit decisioning
- 6 figuresVendor cost avoided, USDThe requirements, architecture and rollout plan for an AI model running on a credit bureau's own hardware, and the vendor review that kept the bureau out of the wrong contract.Credit bureau
Outside lendingSame method, other industries. 8 cases.
- 8 daysFrom the first message to a working version in the founder's handsA Telegram assistant that gets to know each member of a founders' community and introduces them to the people they should meet, modelled on how Boardy works.Business community · Singapore
- 3 daysper website build, down from about three weeksA data centre brand site, plus the AI engineers that made itData centre and infrastructure · Australia
- 29 of 29staff surveyed, free text on every required questionTwo weeks inside a thirty-person agency, priced on their own numbersMedia and creative · Australia
- S$370Ksaved every yearFive departments, one way of working, inside a global manufacturerConsumer goods · Switzerland, head office
- 114,781impressions in thirty daysThe posting system I built for myself, running every day since FebruaryAI consulting · Singapore
- A$26,915booked revenue, in 12 daysOne agent, running the bookings for a hospitality operatorShort-stay accommodation · Australia
- 100%of invoice lines matched to an agreed priceSix suppliers at a Swedish importer, each line matched to the price that was agreedManufacturing and supply · Sweden
- 10 hoursfrom nothing to a working proof of conceptThe engine behind an AI material specifier for architectsArchitecture and sustainability data · Australia