Confidential client · Technical Co-Founder
I came into this as technical co-founder, not as a supplier: the product decisions and the architecture were mine to make. Thousands of records were being posted as ordinary sentences in chat, and nobody could say with confidence what they added up to. Then the platform had to start pricing outcomes better than the tool the market already trusted.
Platform counts, August 2026. Client, sector and product withheld at the client’s request.
The situation
Users were submitting entries as free text in a chat platform. Interpreting them, attributing them to the right person and reconciling the totals was manual, which meant the headline numbers the community relied on were unauditable.
The second problem was sharper. The third-party tool the team depended on for pricing was badly miscalibrated: on some measures it was an order of magnitude away from what the wider market said. Acting on its numbers lost money, and nobody could inspect it to find out why.
My role
Technical co-founder. Idea to production, including the decisions about what not to build.
There was no backlog to work through. The product didn’t exist, and a large part of the job was deciding what it should be: which problems were worth solving, which were interesting but commercially irrelevant, and in what order the remainder should be built.
From there it was mine end to end: the architecture, the parsing pipeline, the pricing model, the backtesting that validates it, the AI layer and the tenancy boundary that keeps clients apart.
This is the engagement shape that’s furthest from outsourced development. Nobody handed me a specification, because producing the specification was the first piece of work.
The outcome
The platform’s value depends entirely on whether its figures can be trusted, so the outcomes worth stating are about auditability and not volume.
What changed
The record of what had happened lived as ordinary sentences typed into a chat tool, so nobody could say with confidence what it added up to. Over 79,000 entries now sit in a ledger that can be audited line by line.
History can be replayed blind against what the market said at the time, so the team knows whether to believe the model when it disagrees. Almost nobody builds this, and it’s the difference between a model you trust and a model you hope about.
A general model can describe this domain fluently and can’t price it. Learning from what happened is the step that moves an answer from plausible to usable, and it’s the step most projects skip. How that model works is the client’s product and stays with them.
The relevant history is supplied to the model at the moment of the question, so a conclusion can be traced, cited and corrected. Anything baked into weights can’t be, which matters when somebody is about to act on the answer with money.
Every call is logged with its own cost. Over 42,000 calls and 433 million tokens have gone through it across four distinct workloads, split between two providers so no single vendor is a dependency.
Answers are checked against the underlying record to a penny on value and 0.1% on rate. A run that reports a gain where the ledger shows a loss fails outright. It doesn’t warn.
The boundary between tenants was treated as a security surface and attacked before it was trusted, rather than reviewed by reading it.
Decisions worth explaining
A language model will happily produce a confident number. Binding its output to the ledger within a penny is what makes it safe to use somewhere the user is about to act on it financially.
The incumbent tool failed because nobody could inspect or check it. Independent providers plus a backtesting engine means the team can see why a number is what it is, and test it before trusting it.
Tenant isolation, state-change gating and privilege changes were reviewed adversarially instead of read through, and what that found was fixed before launch. That’s a code review practice, not a security certification. Where a client needs assurance and not diligence, the answer is an accredited penetration test by people who do only that, and I’ll say so.
Stack. C# / .NET, MongoDB, chat-platform bot integrations, third-party data providers, a fine-tuned model with a retrieval layer, and a hosted model for the analysis layer.
“Ashley has taken our product from an idea into a real, scalable software platform. He’s been able to translate what we wanted to achieve commercially into the product, architecture and technology needed to make it work, while personally leading the development and AI side of the platform. Having someone who can think like a technical co-founder but still be completely hands-on with the engineering has allowed us to move incredibly quickly without compromising on where we want the product to go.”
Work with me
Tell me the problem you’re trying to solve.