Back to home

Things I've built

20 projects across three buckets: the insurance scoring platform I co-founded, products under my own company, and the small tools I wrote because something annoyed me.

Where a model has not been trained against real outcomes, it says so underneath. I would rather you trust the labeled ones.

Co-Founder & Chief Product Officer, RiskLync

Insurance Scoring & Decisioning

A platform that scores insurance leads before and after they are bought, and a marketplace that bids on them in real time. Ten services running on Cloud Run, plus the portal that operates them. I own product strategy and also write a good share of the code.

ZipLync

In production

Pre-bid lead scoring from a phone number and a ZIP code, with no personal data required.

Scores inside the ping so a buyer can decide before paying for the lead. 0.6437 AUC, with an 88x conversion differential between the best and worst decile.

XGBoostPythonCloud RunBigQuery

ProLync

In production

Post-acquisition scoring suite across Medicare Advantage, DSNP, CSNP, and SNP.

Runs after the lead is in hand, with identity resolution against a consumer file. The strongest model in the suite reaches 0.776 AUC.

XGBoostIdentity resolutionCloud Run

TrustLync

In production

Fraud and identity scoring for Medicare leads.

Velocity signals plus identity checks, evaluated at ping time. Works standalone or as an add-on to a scoring call.

PythonFirestoreCloud Run

BidLync

Deployed

Real-time bidding marketplace: auction logic, budget pacing, publisher quality scoring.

Ping to bid under 250ms at P99, including scoring, campaign filtering across fifteen criteria, and the auction itself. Budget is reserved and confirmed through atomic transactions.

Next.jsTypeScriptFirestore

ShieldLync

Deployed

Book-protection and retention scoring for Medicare and ACA books of business.

Finds members at risk of leaving a book before the enrollment window, and ranks who is worth a call.

PythonCloud RunCMS data

GeoRisk

Deployed

Pre-bid risk scoring for property & casualty homeowners leads.

82 features across seven federal and public sources, precomputed per ZIP so the lookup returns in under a millisecond.

A heuristic composite. Not yet trained against outcome data, so it carries no claimed AUC.

PythonBigQueryFEMA / NOAA / Census

AutoLink

Deployed

Auto insurance lead scoring built on street design and crash risk.

Heuristic, same as GeoRisk. Lift is still unproven.

PythonBigQueryCloud Run

SolarScore

Deployed

Solar lead scoring across ten data sources.

Untrained. Built on public data, awaiting outcome labels.

PythonBigQuery

Final Expense EV

Deployed

An expected-value model that prices a lead as conversion times persistence times margin, with a product router on top.

Untrained, with no disposition data yet. The structure is sound; the numbers are not yet earned.

PythonBigQueryCloud Run

RiskLync Portal

In production

The operating platform: campaign management, auctions, reporting, dashboards.

Next.js on Firebase, with a 40-metric pivot engine over transactions and an LLM agent that manages campaigns on a cycle.

Next.js 15FirebaseGemini
Visit site

API Tester

Deployed

A scenario runner for the scoring APIs: 25 test scenarios across six groups, with financial simulation.

Next.jsTypeScript

Owner, BurntOrange LLC

BurntOrange Products

Things I build under my own company, mostly to find out whether I can. Two are live and have real users.

Sports-o-Holic

Live

A multiplayer sports prediction game. Call the next play, score points, talk trash.

Eighteen NPC opponents with distinct personalities, powerups, round modifiers, and a comeback mechanic. Pixel art generated through a prompt-to-sprite pipeline I wrote.

Firebase RTDBVanilla JSImagen
Visit site

AI Legal Assistant

Live

Document management with AI-assisted review for people without a lawyer on retainer.

Next.jsGenkitFirebaseStripeCapacitor
Visit site

Digital Product Catalog

Live

A few hundred digital products across Etsy and Gumroad, with the listing pipeline automated end to end.

Generation, mockups, listing copy, and publishing all run through scripts against the marketplace APIs.

PythonEtsy APIGumroad

Fastbreak

In development

Basketball practice planning for coaches: drill library, practice builder, court diagram editor.

SwiftUIiPhone / iPadFirebase

This Site

Live

Static Next.js export on Firebase Hosting, because the marketing site does not need a server.

Next.js 16Tailwind 4

Personal builds

Tools & Experiments

Small things that solve a problem I actually had. None of them are products, and that is the point.

Meeting Copilot

Deployed

A local meeting recorder with a coaching layer, built to run free by default.

Rules-based coaching runs locally so there is no API bill. The paid model path exists but stays off unless you ask for it.

PythonLocal transcription

Chatlog Archiver

Deployed

Files AI coding sessions into the right project folder automatically, by reading the session title.

NodeSession hooks

Pixel Art Generator

Deployed

Prompt to sprite: generates an image, maps it to a fixed palette, and emits a 24x24 array ready to drop into the game.

NodeImagen

3D Print Profile Tuning

Deployed

Hand-tuned slicer profiles for a QIDI X-Max3, edited as JSON rather than through the UI.

OrcaSlicerJSON

Not listed

Client engagements are described elsewhere on this site without names or identifying detail, because the results belong to the companies that paid for them.

There is also a handful of personal builds I keep to myself: automation for things in my own life that nobody else needs a login to. Happy to talk through any of it in a conversation, just not on a public page.

Want something like one of these?

Most of what I build starts as somebody describing a problem they have been working around for a year.

Start a Conversation