I'm Kyle Bailey, a product leader who ships financial-services platforms. This is where I build working prototypes and frontier-AI projects you can actually click — agents, evals, retrieval, and the governance that makes them deployable — in financial services and beyond.
An original reconstruction of a progressive-commitment onboarding pattern: fetch-and-confirm identity, risk-tiered verification, source-of-funds screening, and a full business path with a party×account authority grid. Flip on “Design notes” to see the product principle behind every screen. Illustrative — fictional bank, no real data.
Open the demo →A multi-agent system that spreads borrower financials and drafts a commercial credit memo with cited figures — reasoning and arithmetic kept strictly separate, graded by a trajectory eval harness.
The instrumentation that tells you whether an LLM feature is getting better or worse: versioned golden sets, LLM-as-judge, trace capture, and a CI gate that fails a PR on regression.
Credit decisions that can defend themselves to a regulator: an interpretable model, reason codes drawn from the model that decided, adverse-action notices, and automated disparate-impact testing.
Point it at a project's dependencies and it tells engineers what breaks, what's deprecated, and what to act on across the version gap — multi-hop GraphRAG with cited sources and honest abstention.
A customer-support agent hardened enough to put in front of the public — prompt-injection defense, capability-scoping, human escalation, and a public red-team report that proves it.
Each ships with a live demo, a public repo, and a written case study — problem, metric, trade-offs.
Eleven years building digital origination and onboarding platforms in banking — most recently leading product for a commercial loan-origination platform. These projects pair that domain depth with the frontier-AI skills enterprise teams actually screen for: not chatbots, but agents, evaluation harnesses, retrieval, and the governance that lets a regulated institution ship them.
Everything here is built on public or synthetic data. Demos are illustrative and not affiliated with any employer's products.