Doing good with AI.

SoYi App — applying Yi Method

Our first product. An emotional wellbeing companion for people who reach for food, cigarettes, or alcohol when something deeper is unmet. SoYi meets you at the moment of impulse, helps you understand what's actually driving it, and guides you toward what you genuinely need. Warm, non-judgmental, grounded in mindfulness and Daoist wisdom.

Fionaa — loan processing agent

An AI agent for automating loan processing workflows. See a simulated run of Fionaa in action.

Building an AI demo can be fast, but putting it reliably into production takes far longer, with no guarantee of success at the end — a rate of failure that far exceeds traditional software engineering projects.

The stats back this up. McKinsey found 88% of companies use AI somewhere, but only 23% are scaling AI agents in even one function. Gartner expects 40%+ of agent projects to be cancelled by 2027. And it's rarely the tech that kills them. It's picking the wrong problem, patchy data, no governance, and costs that creep upwards.

Our co-founder Steve has lived this. He built a quick prototype for an AI Engineering bootcamp he was taking, based on a real lending problem he'd met in financial services. It worked fine for the course. It also skipped most of what makes enterprise deployment hard. So this summer he rebuilt it properly, using the lessons from past projects learned putting AI into production at a Fintech Unicorn. The result is Fionaa. It's not a product we sell. It's how we show what GoodAI can do.

The problem is this: underwriters lose hours to routine checks on business loan applications. Chasing documents, verifying the company, reading policy, cross-checking numbers. Many of those applications never convert into a loan origination, and slow decisions push good applicants to a faster lender.

Here's what Fionaa does with each application, in the order a good underwriter would:

• Are the documents all there? If not, it goes back to the applicant straight away.
• Does the company really exist? It checks Companies House.
• Does the application meet the lender's policy? It goes through clause by clause.
• Do the numbers stack up? It reconciles the application with the accounts and bank statements, and works out whether the repayments are affordable.
• Does the company have a credible online footprint?

The underwriter gets one tidy summary with the evidence attached, and makes the call. Fionaa does the heavy lifting, but with trust and auditability built in.

1. Every execution is recorded. If someone asks “why did it say that?”, you can replay the decision making.
2. We don't take AI's word for it – we verify. Its conclusions are checked against the lender's rules with formal logic-based verification.
3. People stay in charge. The AI accelerates the tedious, routine parts. Judgement stays with your experts.

A case study published by a major UK retail bank took this kind of review from around an hour pre-automation to about two minutes, post-automation.

Fionaa runs on simulated data (see the simulated run). Steve is writing a 6-part series on taking agentic AI from prototype to production; part 1 is on Medium (or read it on LinkedIn if you don't have Medium access).

If you've got a slow, rules-heavy process eating your team's week, we'd love to talk about it.