Sara Huang
Sara Huang

Product Manager · 0-to-1 Builder · New York

Sara Huang

Welcome to my personal portfolio! 😊 I'm a product manager at Lyft in New York, and on my own time I design and build products end to end on the current AI stack. A few of them are below.

Selected work 03 projects
01 / ReList

ReList

Cross-listing tool for resale sellers.

Selling secondhand means re-entering the same item across three marketplaces, about 20 minutes per listing.

Give ReList a photo and a description and it generates a publish-ready listing, recommends pricing from comparable active listings, and sources retail product images to raise listing quality. A per-item chore becomes a single action.

Claude Code Agentic listing generation Active-listing pricing Image sourcing
ReList My Listings: real inventory across eBay, Depop, and Poshmark
ReList price research: comps aggregated across Poshmark, eBay, and Depop with median resale and retail pricing
02 / FitCue

FitCue

Voice-coached workout app.

Written workout plans leave you manually tracking pacing and remembering what to do next.

FitCue turns a prescribed routine into a real-time coached session with spoken cues for pacing, form, and transitions, and keeps your music playing in the background. Give it a physical therapy plan, a coach's program, or any written workout and let it guide you.

Claude Code Real-time voice Eval-driven cue-timing harness
FitCue home: paste a workout and generate a class
FitCue class overview with plan and start
FitCue live coached session timer
03 / Dot

Dot

Personal shopping agent for resale and retail.

Good finds get buried and bad ones waste your time.

Dot watches marketplaces for want-list items and applies encoded buying judgment to every find: sizing, quality, price ceilings, red flags. It notifies only on strong matches. Every verdict is explained, and uncertain cases are escalated rather than guessed.

Claude Code Agent orchestration Human-in-the-loop judgment Escalation logic
In development Demo coming soon
Dot confirm-item screen: deterministic hard filters and Claude-judged taste criteria, nothing watched until you confirm
Dot review queue: graded matches with plain-language reasoning and approve, snooze, or reject actions