I turn fuzzy problems into products people genuinely love.

The problem

Lawyers know the law cold. What they rarely get is practice. Senior attorneys have no time to run mock trials. Existing legal AI tools help with research and drafting. None help you rehearse how you'll actually sound in court.

What I did

We had one day. 10:30 AM to 5 PM. Our team of four talked through lawyer pain points and landed on a sharp insight: this isn't about learning more law. It's about practicing the conversation.

We designed a simple flow. Upload your case. Donna reads it and becomes the other side. She questions you, pushes back, and spots gaps in your logic. Then she gives feedback on clarity, tone, and confidence. We built the live voice experience on Vapi.

The problem

We record every meeting. We promise to read the summary. We never do. Text recaps pile up, and people fake-nod through the next meeting.

What I did

Our team, The No Shows, flipped the format. Instead of reading, you watch. I pushed for a calendar-first experience that feels like Google Calendar, so there's nothing new to learn.

Miss a meeting, open your calendar, hit “Watch Recap.” You're caught up in 60 seconds. Still curious? Ask the built-in agent. It remembers the full transcript. We stitched four sponsor tools into one product and pitched it live to YC founders. No slides to hide behind.

The problem

Engineering leaders can't see which teams are quietly struggling. They rely on spreadsheets and quarterly reviews. By the time trouble shows up, people are already burned out.

What I did

I borrowed a proven idea from behavioral science. Opower cut home energy use by privately showing people how they compared to similar neighbors. We applied that to teams. Each team becomes a village:

• Buildings show capability coverage.
• Gardens wilt when knowledge sits with one person.
• Roads show collaboration with other teams.
• Weather shows workload. Storms mean crunch time.

A scoring engine blends five real signals: cycle time, sprint completion, review speed, after-hours commits, and meeting load. An early warning fires when a team's week slips well below its recent average. AI adds coaching on top, based on what similar healthy teams did differently. Comparisons stay private. Never a public leaderboard. And no one leaving the company ever becomes a game visual.

The problem

Inside big companies, the people with questions aren't the people who write SQL. Every simple question becomes a ticket and a wait.

What I did

I co-led product for AskData, a GenAI analytics platform from Bosch Research. You type a question. An AI agent turns it into a database query, runs the analysis, and draws an interactive chart.

Our goal this semester was to make it ready for internal use at Bosch. I turned a big, fuzzy scope into a clear 7-week plan. Three milestones. Six one-week sprints. One buffer week. I kept a five-person team of engineers aligned with our Bosch sponsor every week. Priorities: a smarter text-to-SQL agent, richer visualizations, secure company login, and ready-to-use demo dashboards.

The problem

Students aged 18 to 24 apply to hundreds of jobs. They bounce between LinkedIn, company sites, and clunky application portals. Conversion is painfully low.

What I did

Our first guess was that students needed mentors. So we went and asked. We interviewed undergrads, grad students, and PhDs. The pattern was clear. They already had advice. What they lacked was help doing the work.

So we pivoted from guidance to execution: one tool that brings matching and applying into a single flow. I used RICE to decide what to build first and wrote the full PRD.

The problem

Recycling breaks at the first step. Nobody can reliably tell what type of plastic something is, how old it is, or if it's recyclable. That gap hurts everyone downstream.

What I did

Our team designed a computer vision system that identifies plastic type, age, condition, and recyclability. We found one root problem shared by three very different customers, and built a fit for each:

• Collectors: a mobile scanning app.
• Manufacturers: a tray-based verification station.
• Cities: drone-based monitoring.

We started with a California-only focus. Then we pivoted to a global solution with all three segments treated as equals.

The problem

Juggling grad school, an internship, and life meant ending each day trying to remember what I actually did.

What I did

I built Work It, a personal note-taking and tracking app. It connects to OpenCode, an AI coding agent. As I work and talk through my day, it writes my daily task log and summary on its own.

The problem

Datadog grew fast in a crowded market. I wanted to know why. And whether it could last.

What I did

I worked through seven years of financials, straight from SEC filings. I built a custom financial model and tracked the SaaS metrics that matter. Then I used the Balanced Scorecard and strategy frameworks to connect the numbers to Datadog's competitive edge.

The problem

Artists need a place to sell work and host events. Fans need a place to find them. That takes many moving parts working together.

What I did

I mapped the whole platform into 12+ services across five layers: data, business, workflow, gateway, and utility. I specced each one and how they talk to each other. That included secure sign-in with one-time passcodes.