Projects
At work, I've built agents for bug filing, PRD generation, and user interview summarization. This space highlights what I build for fun. Both of these started as personal frustrations and turned into products I actually use, and how I learned to build end-to-end, from scoping and design through to deployment.
Web App · Generative AI · React
FindMySki
An AI-powered ski recommender that walks you through a short intake, then explains exactly which ski to buy and why, drawing on curated expert reviews and retailer data
Demo
The Problem
As someone who skis a lot, I kept getting asked by friends what ski to buy. I'd give an opinion but realized I was just guessing based on my own experience. Ski selection is genuinely complex. It depends on ability level, skiing style, body type, terrain preference, and budget. Existing tools either oversimplify to a few dropdowns or dump a spec sheet on you with no guidance on what actually matters.
What I Built
A web app that walks you through a short intake about your ability, goals, body type, and budget. It uses RAG to pull from a curated knowledge base of expert ski reviews (OutdoorGearLab, SKI Magazine, Blister, Switchback Travel, and more), then uses an LLM to synthesize a personalized recommendation with a plain-English explanation. I intentionally made the reasoning more prominent than the recommendation itself, because that's what actually helps someone feel confident in the purchase. Results can be shared via link or emailed directly.
PM Instinct Behind It
"When you're the workaround people use because the real tool doesn't exist, that's a product gap worth filling. The key insight from talking to people was that they didn't distrust the recommendation, they just didn't know how to evaluate it. Explaining the 'why' in plain language is what actually builds confidence, so I made it the centerpiece rather than an afterthought."
Web App · NLP · Mapping · Cycling
RideRouta
A ride planning tool that lets road cyclists create routes by describing their ideal ride in plain English or using a form for more control
Demo
The Problem
Strava and most GPX tools optimize for distance and connectivity, not ride quality. For road biking specifically, that means you regularly get routed up punishing climbs you didn't ask for, through intersections with heavy car traffic, or onto roads with no shoulder. Every time I wanted to explore a new route around SF I'd spend 30+ minutes in Strava manually adjusting, cross-referencing Google Maps, and checking Street View just to sanity-check the road conditions. It was a whole process before I even got on the bike.
What I Built
A web app where you describe your ideal ride in plain English or use a form for more control, and it generates a route optimized for road cycling on lower traffic streets with manageable elevation. It connects directly to Strava, exports to GPX for Garmin, and shows elevation profiles so you know what you're getting into before you clip in. The whole point was to eliminate the 30-minute pre-ride planning ritual and just get out the door.
PM Instinct Behind It
"The temptation was to build a general cycling tool, but I knew from my own experience that road biking has very specific needs that are different from mountain biking or commuting. Scoping it tightly to one persona in one geography made the product actually useful rather than mediocre for everyone. I also learned a lot about the tradeoffs between routing APIs, elevation data sources, and what 'low traffic' actually means when you try to operationalize it."
Let's build something
Have an idea you want to collab on?
linkedin.com/in/vivfeng
