Portfolio

I have shipped production systems across commerce, AI, and decentralized finance — from five-person startups to Fortune 500 deployments. These are the systems behind the methods on my How I build page. Each was architected, not just assembled.

Tekel — a barter economy, in software

Co-founder & CTO · token-mediated marketplace and protocol

Screenshots

ChallengeBarter breaks down the moment two parties value a trade differently — someone always comes up short, and the deal dies. Solving that normally means reintroducing a central currency and the dependencies that come with it. The goal was a marketplace where unequal trades still settle fairly, without a bank in the middle.

ApproachWe designed the economic mechanism, not just the app: a token that absorbs the imbalance in any trade, priced by an oracle built from the network's own trade data, with credit limits and a non-inflationary treasury funded by transaction fees rather than printing supply. On top of that protocol we built the full marketplace end to end — Stripe Marketplace for credit card processing and merchant payouts, escrowed payments, multi-party dispute resolution, milestone and time-and-materials billing, and five fulfillment models — across web, iOS, Android, merchant, and admin surfaces.

ResultA complete multi-sided commerce platform run by a one-person engineering org, on edge infrastructure at near-zero fixed cost — the trust mechanics (escrow, disputes, phased payments) that normally require dedicated payments and operations teams, delivered as a single coherent system.

Economic mechanism design

Edge-native

TypeScript

Stripe

Multi-platform

Goes deep:

Architecture

·Type-safe contracts

·Error handling

Masterypath — rebuilt for the AI era

Founder · AI-native learning platform

Screenshots

ChallengeA learning platform built in 2020 — mind maps reinforced by spaced repetition — was sound in concept but carried the operational weight of a self-managed Kubernetes stack, and predated the tools that would let AI do real work inside it. The question was not how to patch it; it was whether to rebuild it.

ApproachWe made the harder call: a ground-up rewrite. The self-managed cluster was retired for a serverless edge architecture that eliminated nearly all fixed infrastructure cost and operational burden. AI was built in as a content-creation engine rather than a bolted-on chatbot — organization-scoped agents with private knowledge bases, retrieval-augmented answers that cite their sources, and pipelines that turn raw documents into structured lessons and auto-generated flashcards. The mind-mapping engine was extracted as a standalone, independently published library.

ResultA second-generation product that does more at a fraction of the running cost of the first — and a reusable visualization library with its own release lifecycle, an asset beyond the application itself. A demonstration that the right architectural decision is sometimes to discard your own prior work when the economics change.

RAG

AI agents

Serverless/edge

Reusable libraries

TypeScript

Goes deep:

Type-safe contracts

·Automated delivery

·Documented decisions

Municipal AI assistant — answers and acts

Principal Engineer (contract) · multi-tenant RAG platform for municipalities

Screenshots

ChallengeLocal governments field the same citizen questions endlessly, and the answers live scattered across their own websites and PDFs. A municipal SaaS provider needed a white-label AI assistant each city could brand and deploy against its own content — accurate enough for government use, and able to do more than just answer.

ApproachI built the core product end to end. The retrieval layer ingests a municipality's sites and documents, chunks and embeds them, and answers questions with citations back to the source — tuned for public-sector accuracy, with a debug pipeline that traces every answer to its origin. Then I took it past Q&A: a function-calling agent that files 311-style service requests conversationally, integrating with each city's existing systems, with autonomy configurable from fully automated to human-confirmed. The platform was engineered for SaaS economics — per-tenant isolation, token budgets, and admin tooling so non-engineers can onboard new tenants.

ResultA deployed multi-tenant platform that became the client's core offering — assistants that do not just inform citizens but complete transactions for them, reducing support load while keeping every answer auditable back to its source.

RAG

Function-calling agents

Multi-tenancy

AWS

Vector search

Goes deep:

Architecture

·Error handling

·Testing

Lucra — a stablecoin that never liquidates

Co-founder · audited DeFi protocol on SolanaTry the live demo

Screenshots

ChallengeMost collateralized stablecoins protect their peg by liquidating users' collateral when markets move — punishing the very people who back the system. The goal was a fully decentralized, over-collateralized stablecoin that holds its peg without ever seizing customer collateral, in difficult market conditions, with no room for error in the math.

ApproachCo-leading a five-person team, we designed the protocol's economics from first principles — custom on-chain oracle, bond, and lending systems, with protocol-driven arbitrage doing the peg enforcement that liquidation usually does. Eighteen months of research, mathematical modeling, and rigorous testing produced the full stack: audited smart contracts, the on-chain programs, a typed Web3 SDK, a React interface, and CLI tooling.

ResultIndependently audited contracts delivered — a 300%-collateralized stablecoin with no liquidation of customer collateral, the harder and more user-aligned design, shipped despite a hostile market.

Smart contracts

Solana/Rust

Protocol design

Oracles

Web3 SDK

Goes deep:

Architecture

·Documented decisions

Before the recent work — two decades of shipping at scale.

Enterprise AI & automationAt Amelia, built ML classification for client event analysis tied to $27M in new revenue and configuration management that let a handful of engineers operate 100K+ production devices for Fortune 500 clients including The Home Depot. At Cisco, led ten engineers raising autonomic incident remediation to ~80%, saving $8.8M annually.

IPncm — large-scale network configuration managementAt IPsoft, designed IPncm after a managed-services SOW required upgrading every IOS device in a customer production environment within 4 hours of any reported security vulnerability — a bar set using the Cisco vulnerability detection framework I had already built. The platform upgraded 30,000+ Cisco IOS devices inside that window with network-test rollback guarantees, peaked at 200,000+ devices under management over a nine-year run, and expanded to all major network vendors plus Linux over SSH. Small teams used it to replace 15–50 FTEs depending on the year; it anchored a $75M annual contract that ran for five years. Source on GitHub.

Mission-critical systems under pressureAs a U.S. Army Warrant Officer, deployed tactical communications across 150+ sites in Afghanistan — command and control for 75% of U.S. Army tactical operations in-country, with 3,000+ routing devices and 25,000 users under RC South. Shipped custom web-based IP management software under sustained combat conditions. Bronze Star for meritorious service.

First company, founder to exitCo-founded Solern from self-taught web development into a market leader in two categories of hospitality and gaming software — casino surveillance, encrypted communications, facial recognition, and HRIS/ATS with a job board that reached 500K members. Grew the business to 400+ corporate clients, 500K+ users, and $1.5M annual revenue, culminating in a successful IP sale.