Product walkthrough
The journey end-to-end: installer voice intake, customer phone journey, audit dashboard. Start here.
Lending Agent
An AI-mediated agentic credit broking demo. Plain-English customer journey, structured disclosures and consents, sequential lender waterfall, replayable audit log.
Quick links by audience
Section titled “Quick links by audience” For retailers If you sell solar, batteries, EVs, or any high-value home-improvement that customers finance, start here.
For brokers If you hold the FCA permission and arrange consumer credit through a panel of lenders.
For lenders If you sit on a broker's panel and want to understand what changes when the journey is agent-mediated.
For regulators & risk teams Consumer Duty, CONC, FCA AI strategy, vulnerable customers, replay-as-evidence.
See also: the Lending Agent family Three sibling demos. Lending Agent (waterfall, this site), Lending Agent Presenter (menu-style finance presentment), Lending Agent Oversight (AR network supervision for principal firms).
What you’ll find here
Section titled “What you’ll find here”Architecture
State machine, reconciliation, event protocol, cold-start recovery, mock-vs-real boundary. Read the architecture.
Safety
Threat model, prompt injection, hallucination defences, fail-safe state machine, model evaluations. Safety overview.
Privacy
UK GDPR, DPIA framework, data minimisation, retention, sub-processors, PECR. Privacy overview.
Regulatory
Consumer Duty, CONC, FCA AI strategy, vulnerable customers, replay-as-evidence. Regulatory overview.
Deploy your own
Run locally, deploy to Vercel, wire a custom domain, harden for production. Deploy guide.
The TL;DR
Section titled “The TL;DR”A working prototype that takes a UK consumer-credit journey from doorstep to lender decision through an AI agent. The journey is a sequence of regulated gates (status disclosure, eligibility, indicative quote, application details, vulnerability check, credit search consent, pre-contract, sequential lender waterfall, decision). Every gate produces structured evidence. Every disclosure is replayable for compliance scoring. Customers can withdraw at any time.
The model is a narrator over a state machine. Regulated content never comes from the model. The deterministic side recovers from any model confusion automatically.
This is a personal side project, built independently of any firm. The journey, audit, and replay layers are real. The lender panel, decisions, and credit search are mocked. Documentation is for productisation conversations.