Work

Building the world's most connected recruiter.

• Designed and shipped Candidate Outreach, the system that owns every proactive message Fonzi sends a candidate. Features open items fire-and-forget; one per-candidate Temporal entity workflow decides timing, bundling, contact budget, staleness, and channel. Consolidated nine independently scheduled notification types onto it, 18 scenarios live.
• Killed a class of failure where the agent sent its own internal deliberation to real candidates. Built a byte-exact replay harness over recorded production traces plus a live-loop instrument, ran a five-arm bake-off across roughly 2,700 scored trials, and let the measurement pick the architecture. The winning design routes every outgoing message through a send tool: 27 leaks in 116 runs down to 0, with voice scored unchanged.
• Gave the agent a second channel after measuring that 349 outreach items across 261 candidates died unreachable in 28 days, and that 30% of outreach-active candidates had no text thread at all. Built email end to end: schema, webhooks, per-Interview-Request threading, SPF/DKIM/DMARC, one-click unsubscribe, and suppression. Retired six Customer.io templates behind it.
• Built the evaluation practice for a probabilistic product. Judges extract and code grades, which took judge accuracy from 4/9 to 9/9 on adjudicated production replies. Established that replay fidelity is an interaction effect: local tool definitions alone reproduced a failure 19% of the time and message flattening alone 6%, but both together 75%.
• Made interview times correct across timezones after a Pacific candidate got a 4:45am text. Modeled candidate timezone as first-class data with provenance, replaced a fingerprint heuristic that silently misread evening-Pacific interviews as date-only, and ran the migration as 16 stacked changes. Verified in production that one scenario released at 13:00Z for New York candidates and 16:00Z for San Francisco.

Seed to Series B.

• Owned Tap to Pay on iPhone end to end: refactored the checkout backend to support custom cover charges rather than fixed tickets, integrated Stripe Terminal with React Native and Expo, and ran beta testing with the success team. $1.5M net new revenue and $10M in total sales in the first year.
• Re-architected the event discovery backend on MongoDB materialized views and background tasks to support the mobile app redesign. Cut p95 latency 67% (2s to 650ms), absorbed 1.75M requests on Halloweekend 2025, and contributed to a 64% increase in marketplace conversion.
• Owned two critical production incidents during the Next.js migration: traced 502s to a load balancer and ECS misconfiguration, and a memory leak to an incorrectly server-rendered React Query client.
• Found that a team of 10 was waiting 20+ minutes to merge, then shipped a merge queue that removed the wait and kept semantic merge conflicts out of production.
• Set up a weekly mobile release process with the engineering manager, adding structured QA checkpoints that caught bugs before production.
• Reviewed 33% more code than the next engineer on the team.
• Cut deploy times 25% by moving the pipeline onto Turborepo.

• Led a zero-downtime migration of web checkout to tRPC while it processed 400 to 500k orders a month, roughly $1M a week, adding test coverage and aligning endpoints with the mobile client.
• Built the backend APIs and redesigned the web frontend for event discovery, with filters for time, price, size, and neighborhood. Drove around 2,000 monthly orders at 10% explore-to-purchase conversion.
• Cut CI/CD build time and cost 50 to 60% with local test containers, SWC, parallel jobs, and trimmed lint dependencies.
• Championed Graphite for stacked pull requests across a team of 7. Adopters merged 200 to 300% more code than non-adopters.

• Integrated the GitHub API into a standup visibility tool used by 3 client teams, surfacing PR reviews, Linear statuses, and calendar events to unblock engineers.
• Built the authentication flow and CRM for a seed-stage experiential members club on MongoDB, Express, and Next.js.
• Reviewed code for 3 engineers and introduced annotation and description standards that made reviews easier to follow.