Senior Product Manager
Operator-grade product leadership across regulated SaaS, marketplace, and ecommerce. I ship product that moves the metric.
B2B SaaS · Marketplace · AI-Enabled Product
About
I started in mortgage finance and secondary marketplaces, and found I liked building systems that help people use a company's resources more efficiently. I prototyped with spreadsheets and slideshows before I knew "product manager" was a role. I care about products that move outcomes, where I can point at the metric and say that's what we changed and how we know. My best work is hands-on, close to the product and close to the customer.
Selected work
Problem. Subscribers needed their crypto activity to land cleanly in the tax tools they already used.
Decision. Mapped a partner API to our data model and ran a two-week live spike before signing, then shipped it as the headline growth driver.
Problem. Growth depended on rented ad traffic with no owned front door for buyers and sellers.
Decision. Converted rented ad traffic into an owned marketplace front door customers came to directly.
Problem. Two distinct audiences needed single-minded messaging without doubling the engineering.
Decision. Stood up two branded front ends on one shared foundation, with messaging tuned per audience.
Problem. Per-conversation AI cost threatened the unit economics of the subscription.
Decision. Re-architected bot context with a per-bot vectorized index so each call only paid for what it needed.
App builds
Describe a company vision and AI grounds it in a market read and a lean-to-premium strategy spectrum. One dial plus a team-size input drive the development cost, timeline, and budget, then a button generates a build-ready PRD. AI runs off the hot path; the board math is a client-side, inspectable model.
Type a vague travel wish and get back a few fully-priced trips: flights, a stay, a car only if needed, affordable highly-rated restaurants, and things to do split into free, under $25, and one splurge. Claude runs live web search once per prompt to ground real destinations and prices; the budget dial, sort, and filters are a client-side engine.
Answer one short questionnaire and see which sustainability regulations apply (EUDR, CBAM, CSRD, CSDDD, PPWR, California SB 253/261), what data each needs, and where one answer satisfies several. Fully client-side, reading a version-controlled, sourced data library with confidence labels and unverified-fact flags. A demonstration, not legal advice.
A hyperlocal peer-to-peer lending community where neighbors lend and borrow physical items. The product bet is a reversed points economy: lending mints points and borrowing spends them, because lenders carry the risk and are the scarce side. Every point movement is an append-only ledger entry settled inside a database transaction, with reputation, tiered access for higher-risk gear, and distance-based discovery. Postgres and Auth.js back it; no real money is involved.
A professional-grade AI copywriter built for working copywriters, not a replacement for one. Upload past work and it models three layers, how you sound, how you build copy, and how you lay it out, then drafts new copy in your voice with awareness-driven frameworks. Files are read in the browser; only text is sent. Streamed output with one-tap iterations, server-side keys, per-IP rate limiting, and graceful degradation when no profile is given.
Feasibility studies for resilient communities, sustainable villages, and workforce housing. Set a population and self-sufficiency target, choose the food, energy, water, and housing systems to build, and get the workforce, land, utilities, and capital it would take, plus an audience-ready feasibility report with a scorecard, verdict, and exportable PDF. The engine is a client-side, inspectable model built on a relational blueprint of 1,940 capabilities across 16 systems.
A glass-box decisioning console that replays synthetic subscriber and audience history across a 26-brand B2B media portfolio and recommends a per-reader paywall action. It routes each reader on firmographics, scores reader archetypes, gates on confidence, then picks harvest or invest, and shows what a static wall actually earned next to what the engine could earn across three value tracks: subscription revenue, enterprise pipeline, and advertiser audience. Deterministic and seeded, every number traces to an inspectable formula, and it runs as a single offline file with no dependencies or network.
A generative geometric pattern studio: seventy-two equation families (Voronoi, flow fields, phyllotaxis, superformula blooms, strange attractors, Lorenz butterflies, Fourier epicycles, orbital choreographies, quasicrystals, harmonographs, metaballs, Julia and Newton fractals, Apollonian gaskets, spirographs, Chladni figures, moiré interference, demoscene plasma, Electric-Sheep-style fractal flames, warp-feedback tunnels, kaleidoscopic Kalisets, gyroid flux, original fields — sine-fold turbulence, vortex braids, Möbius whirlpools, drifting charge fields — line work like torus knots, rotating tesseracts, string-art caustics, and swaying L-system plants, plus domain-warped marble and more) rendered on canvas from a seeded PRNG. Every pattern is a function of its seed, so the URL reproduces it exactly; geometry and color are independent, so harmony-based HSL palettes recolor without touching the structure. Patterns also animate — each family evolves through a native, seamlessly looping motion — with still-image and looping WebM/GIF export. Renders at 2× for crispness and exports from the equations at up to 4K, never upscaled. 100% client-side — no backend, no API, no accounts.
Borrow the commodity, build the differentiator. A full browser image editor built on the open-source miniPaint (MIT, attributed) for the solved parts (canvas, layers, filters, export), extended with the actual product: a natural-language edit layer. Describe an edit in plain words; Claude returns a JSON plan (intent to structured steps) that runs entirely client-side against miniPaint's own filter and transform functions, so every AI edit lands in normal undo. Save plans as reusable recipes that replay with zero API calls, plus an opt-in AI critique. Manual editing is 100% on-device; images never leave the browser except one explicit, downscaled critique snapshot.
Skills & toolkit
Certifications