PORTFOLIO

Selected work

A deliberate lineup, ordered to read as range rather than one trick: the GEO + MCP + evals differentiator, a shipped full-stack product, backend and distributed depth, agentic DevOps across a cross-country team, and a production-reliability war story. Every claim is grounded in the real record — and where something is a foundation, a prototype, or operated rather than built, it says so.

01 —

Selected work

five case studies

live

geocheck

An AI-search brand-visibility (GEO) checker that is both a CLI and an MCP server, shipped with its own reproducible eval harness — transparent unit-tested metrics and an extractor graded against human labels behind a CI gate. Runs offline with no API key.

GEOMCPEVALSPYTHON

Read case study →

live

MapleScholar

A production, nonprofit AI research-discovery PWA — paste a DOI / title / arXiv / OpenAlex ID to resolve a paper across 7 scholarly APIs, chat with it in 70+ languages, and read it as interactive HTML. Sole primary engineer, end-to-end, with a first-class GEO program.

NEXT.JSRAGSUPABASEGEOPWA

Read case study →

CiteStreak

The greenfield 2026 GEO / AI-visibility SaaS I am currently building. The tested platform foundation — RLS tenant isolation, an append-only credits ledger, SELECT FOR UPDATE quota reservation, replay-safe Stripe webhooks, a Temporal refund workflow — landed across Phases 0–3 with 24/24 tests green. The product surface is still in progress.

NESTJSTEMPORALRLSSTRIPEGEO

Read case study →

archived

Guacamoly remote-desktop

A self-hosted, browser-based multi-user remote-desktop platform on AWS EC2 — Apache Guacamole + XRDP/VNC + MariaDB JDBC auth, hardened with a hand-built WireGuard split-tunnel firewall and run via an AI-agent-driven EC2-to-EC2 migration. 5 isolated, audio-enabled sessions across the US, Netherlands, and Kenya on one box.

AWSLINUXNETFILTERAGENTIC-DEVOPS

Read case study →

archived

The bug that made our alerts lie

On a production AI-visibility platform I operate, error alerting had been silently broken for months — a log-prefix bug from the process manager broke level extraction in the ingestion pipeline, so level="error" matched nothing. Found by querying the log store directly; fixed with one pipeline stage; verified via the Loki API.

OBSERVABILITYSREGRAFANALOKILLM-OPS

Read case study →