Projects I’ve Built
Interactive prototypes born from real user research and deep product thinking.
10+ experiments | Agentic workflows, RAG, prompt engineering, full-stack AI products
As a Product Manager, I research and understand pain points, validate with users, and build high-fidelity interactive prototypes. These are the personal projects I've built in the last 12 months using Cursor, Vercel, Claude, LangChain/LangGraph, RAG pipelines, and no-code + code hybrids. Many started as class projects or weekend experiments, and some became full demos I've tested with real users. Scroll down and click "Try the Prototype" on any of them.
Agentic workflows, RAG, prompt engineering, full-stack AI Products
As a Product Manager, I don’t just ideate - I research and understand pain points, validate with users, and build high-fidelity interactive prototypes that bring the full product vision to life. These are the personal projects and prototypes I’ve built in the last 12 months using Cursor, Vercel, Claude, LangChain/LangGraph, RAG pipelines, and no-code + code hybrids. Many started as class projects/ weekend experiments and turned into full demos I’ve tested with real users. Scroll, click “Try the Prototype” and see the prototypes.
NXTai - Decision Intelligence for Product Teams
- Designed a Product Memory Graph that unifies scattered signals - support tickets, surveys, CRM notes, reviews, product backlog and strategy into one contextual knowledge layer.
- Engineered ROI-weighted prioritization that quantifies revenue risk, churn probability, and retention value for every feature request.
- Built closed-loop automation pushing ranked recommendations directly into Jira and Notion, eliminating manual triage cycles.
Feedback Sources Supported
10+
Time to Ranked Backlog
<2 min
NXTai - Decision Intelligence for Product Teams
- Unified support tickets, surveys, CRM notes, reviews, and backlog into one Product Memory Graph.
- Ranked feature requests by ROI, quantifying revenue risk, churn probability, and retention value.
- Pushed ranked recommendations straight into Jira and Notion, cutting manual triage.
Feedback Sources Supported
10+
Time to Ranked Backlog
<2 min
NXTai - Decision Intelligence for Product Teams
- Unified support tickets, surveys, CRM notes, reviews, and backlog into one Product Memory Graph.
- Ranked feature requests by ROI, quantifying revenue risk, churn probability, and retention value.
- Pushed ranked recommendations straight into Jira and Notion, cutting manual triage.
Feedback Sources Supported
10+
Time to Ranked Backlog
<2 min
TalentSphere - Agentic Career Intelligence for New Grads
- Built a multi-agent system mapping a candidate's background to best-fit roles, industries, and skill gaps against live market data.
- Auto-generated tailored applications and recruiter outreach from the match.
- Shipped end-to-end in under 48 hours at the UW MSIM Hackathon, winning Most Innovative and Viable Product.
Time to Career Clarity
<60 sec
Workflows Automated
4
TalentSphere - Agentic Career Intelligence for New Grads
- Built a multi-agent system mapping a candidate's background to best-fit roles, industries, and skill gaps against live market data.
- Auto-generated tailored applications and recruiter outreach from the match.
- Shipped end-to-end in under 48 hours at the UW MSIM Hackathon, winning Most Innovative and Viable Product.
Time to Career Clarity
<60 sec
Workflows Automated
4
TalentSphere - Agentic Career Intelligence for New Grads
- Built a multi-agent system mapping a candidate's background to best-fit roles, industries, and skill gaps against live market data.
- Auto-generated tailored applications and recruiter outreach from the match.
- Shipped end-to-end in under 48 hours at the UW MSIM Hackathon, winning Most Innovative and Viable Product.
Time to Career Clarity
<60 sec
Workflows Automated
4
Rotten Tom-AI-toes - Clinical AI Governance & Trust Scoring Platform
- Built an evaluation wizard scoring models across six trust dimensions, from accuracy and safety to fairness and explainability.
- Benchmarked 17 foundation models on clinical suitability, with trends tracked over time.
- Engineered enterprise-grade dashboards with real-time risk gauges, radar charts, and score rings, inspired by Datadog and Palantir Foundry.
Trust Dimensions
6
Models Benchmarked
17
Rotten Tom-AI-toes - Clinical AI Governance & Trust Scoring Platform
- Built an evaluation wizard scoring models across six trust dimensions, from accuracy and safety to fairness and explainability.
- Benchmarked 17 foundation models on clinical suitability, with trends tracked over time.
- Engineered enterprise-grade dashboards with real-time risk gauges, radar charts, and score rings, inspired by Datadog and Palantir Foundry.
Trust Dimensions
6
Models Benchmarked
17
Rotten Tom-AI-toes - Clinical AI Governance & Trust Scoring Platform
- Built an evaluation wizard scoring models across six trust dimensions, from accuracy and safety to fairness and explainability.
- Benchmarked 17 foundation models on clinical suitability, with trends tracked over time.
- Engineered enterprise-grade dashboards with real-time risk gauges, radar charts, and score rings, inspired by Datadog and Palantir Foundry.
Trust Dimensions
6
Models Benchmarked
17
Kindred - AI-Powered Therapist-Client Matching Platform
- Scored fit across four clinically-grounded dimensions: therapeutic style, primary focus, depth orientation, and identity and cultural fit.
- Interviewed 25+ therapists and clients, finding 80% run unpaid screening and lose $6K-12K a year to mismatched referrals.
- Shipped a working intake prototype and won the eBay Best Marketplace Idea Prize at UW's Dempsey Startup Competition.
Matching Dimensions
4
Revenue Leakage per Therapist
$6K-12K/yr
Kindred - AI-Powered Therapist-Client Matching Platform
- Scored fit across four clinically-grounded dimensions: therapeutic style, primary focus, depth orientation, and identity and cultural fit.
- Interviewed 25+ therapists and clients, finding 80% run unpaid screening and lose $6K-12K a year to mismatched referrals.
- Shipped a working intake prototype and won the eBay Best Marketplace Idea Prize at UW's Dempsey Startup Competition.
Matching Dimensions
4
Revenue Leakage per Therapist
$6K-12K/yr
Kindred - AI-Powered Therapist-Client Matching Platform
- Scored fit across four clinically-grounded dimensions: therapeutic style, primary focus, depth orientation, and identity and cultural fit.
- Interviewed 25+ therapists and clients, finding 80% run unpaid screening and lose $6K-12K a year to mismatched referrals.
- Shipped a working intake prototype and won the eBay Best Marketplace Idea Prize at UW's Dempsey Startup Competition.
Matching Dimensions
4
Revenue Leakage per Therapist
$6K-12K/yr
As a Product Manager, I don’t just ideate - I research and understand pain points, validate with users, and build high-fidelity interactive prototypes that bring the full product vision to life. These are the personal projects and prototypes I’ve built in the last 12 months using Cursor, Vercel, Claude, LangChain/LangGraph, RAG pipelines, and no-code + code hybrids. Many started as class projects/ weekend experiments and turned into full demos I’ve tested with real users. Scroll, click “Try the Prototype” and see the prototypes.