Selected builds and work

What a human can do now

A compliance officer at a Canadian fintech can monitor OSFI, FINTRAC, and CSA simultaneously and receive a structured, reasoned briefing within minutes of a regulatory update being published. Previously this required an analyst to manually read each source, cross-reference updates against internal processes, assess materiality, and write up a recommended action -- a process that requires increasing headcount to maintain as product lines scale. The automated system scales without headcount.

Live example

During development, the system processed a live FINTRAC update on extortion-related money laundering, correctly identified AML Transaction Monitoring as the primary affected workflow, flagged KYC Onboarding as a secondary impact, and caught a non-obvious connection to Data Retention and Privacy through an information-sharing obligation the primary mapping missed.

Read the full case study

The problem

Claims management breaks in a predictable way. The system that worked at 50 claims a month fails at 500 -- not because the team got worse, but because spreadsheets weren't built to track a full claims lifecycle across providers, members, payers, plans, and coverage tiers simultaneously. The result is revenue leaking through gaps nobody owns: denied claims that never get appealed, member balances that age without follow-up, and financial reporting that nobody trusts.

What changed

Denial rate held at 7% against a 12% industry average. Days in AR reduced from 60+ to 43. Approximately $400K in denied or stuck claims recovered over 18 months.

The problem

Staying current on a set of YouTube channels is time-consuming. You either watch everything or miss things entirely.

What I built

A Python bot that checks a configured list of YouTube channels on a schedule, detects new uploads, generates AI summaries of each video, and posts a digest to a Slack channel. Built with full API integration and deployed as a persistent local background worker.

What a human can do now

A real estate broker can drop PDF brochures into a folder from any device and have a client-ready interactive survey website rebuilt automatically within minutes -- without touching a line of code, opening a design tool, or manually entering property data. Previously, building a comparable survey required manually extracting specs from each brochure, formatting a comparison table, and assembling a document by hand.

What it produces

A single HTML file with property cards, a filterable comparison table, a clickable map, per-property feedback and comment tools, and a print layout that produces one formatted page per property. A watcher script monitors a OneDrive folder and rebuilds the site automatically when brochures are added or removed -- no manual trigger required.