Portfolio
A selection of needle-moving products from 20 years of building.
Led the transformation of Roofstock's proprietary property management system to include agentic AI automations — enabling a smaller operational team to support more rental homes. Automations covered invoice approvals, rent price adjustments, maintenance request intake, and delinquency follow-ups. These automations reduced mundane operational work and drove meaningful cost reductions across the business.
Led the transition from outsourced rent collection to in-house payments processing built on Stripe. Internalizing this core operational function eliminated third-party fees and gave the company full control of the payments experience and data.
AirCover guests who'd just had a host cancellation needed help finding an alternative — but the original experience focused on the guest's refund rather than proposing rebooking alternatives. Rebuilt the web and native rebooking experience, and led three generations of rebooking-suggestion ML model iteration: removing price filters, shifting to search-based retrieval, and layering in behavioral signals. Included an LLM-powered feature that summarized a guest's original booking preferences and invited them to refine.
When a host canceled 30+ days before check-in, guests had the worst NPS and lowest rebooking rates because they received the weakest support. Identified the gap through data and agent shadowing: same customer pain, wrong playbook. Extended instant credit eligibility, proactive outreach, and in-app rebooking support to this large customer segment.
Rebooking credits drove higher rebook rates, but blanket incentives were expensive and blunt. Redesigned from fixed-value credits to ML-driven personalized offers based on guest price sensitivity and booking characteristics — targeting guests who needed encouragement most, at the smallest effective offer size.
Hired to build a commercial real estate marketplace for tenant brokers. Six months in, the business shifted to immediate monetization. Led the team through the strategic reset — rebuilt the roadmap around listing broker value, shipped Company Views (de-anonymized tenant engagement data that helped brokers revive stalled deals), and launched marketing automation tools.
Academia had 5M MAU but low per-user engagement — no "what to read next" experience existed for researchers. Started with a hackathon proof-of-concept using TF-IDF applied to the text corpus; iterated through email, feed, and right-rail surfaces. Recommendations drove higher per-session content engagement and re-activation of dormant users.
The original Premium product was built for academic authors — but a significant portion of traffic was researchers and grad students with different Jobs-to-be-Done. When Premium growth stalled, re-anchored on this new user segment, ran tests and customer research to identify the next generation of Premium products.
I'm happy to share more context on any project — or talk through how the work might apply to your current challenge.
Get in touch