These are the ones worth a full write-up. Newer builds usually land on GitHub first. github.com/summerstateofmind
Pulls Seattle's code-violation records, maps every flagged property, and scores each one by distress signal, so investors spot motivated sellers before the listings ever appear.
What It Does
A code violation is a quiet tell that an owner is checked out. FlaggedLeads reads those public records, ties each one back to its owner, and ranks the property by how likely that owner is to sell. Investors get the lead before anyone else knows it exists.
What I Built
Built solo, end to end. An ingestion pipeline cleans messy municipal records and matches each violation to the right parcel; a scoring engine turns volume, severity, and recency into one sortable number and shows its work. An early-warning layer reads 311 calls and permit filings to flag a property before the violation even lands. Owner enrichment catches absentee landlords, PO-box forwarding, and multi-address portfolios.
Tech Stack
Live Now
12,000+ active violations across 9,500+ Seattle properties, refreshed continuously. Free map to browse; a paid tier unlocks scores, owner data, and CSV export.
Compares the true cost of a move across 165 cities, weighing not just rent and salary but childcare, healthcare, vacation, and parental leave. The same engine is open to AI agents over MCP.
What It Does
Most calculators stop at salary and rent. cityparity also prices the social safety net, turning subsidized childcare or universal healthcare into an equivalent salary. The result: "the lower-paying city actually pays more" becomes a number you can show your partner, not a hunch.
What I Built
The financial engine is pure JavaScript. It models progressive tax across 69 countries and runs the same code in the browser and the test suite. Your salary never leaves your device; every calculation is client-side. I also shipped a public MCP server so AI assistants can run the comparisons themselves, plus a programmatic-SEO system that generates a ranked page for every city pair.
Tech Stack
Live Now
165 cities across 69 countries. No account, no tracking. Your numbers never leave your browser. AI agents can connect at mcp.cityparity.com.
resuumi AI
Automated Resume Keyword Optimization
GPT-4 powered tool that ingests a resume and job description, optimizes keyword matching, and outputs an improved, downloadable PDF.
The Problem
Job seekers waste hours manually tailoring resumes for each application, often missing critical keywords that ATS systems filter on. Most don't know which terms matter.
My Role & Approach
Designed the product end-to-end and built the UI in Panel (Python). Integrated GPT-4 for semantic keyword analysis and resume rewriting. Shipped to production as a self-serve web app.
Tech Stack
Status
MVP shipped August 2023
.png)
Spellforge
LLM Quality Evaluation with Synthetic Users
Scalable quality evaluation platform for LLM apps using simulated synthetic users and AI judges.
The Problem
LLM-powered apps are difficult to test at scale. Manual QA is slow and subjective. Teams need automated, repeatable evaluation that simulates real user behavior.
My Role & Approach
Conceived and built the platform. Designed the synthetic user simulation framework and integrated AI judge evaluation. Built the Streamlit interface for team workflows.
Tech Stack
Status
MVP shipped October 2023
.png)
.png)
DoppelFlow
AI Agents for Desktop & Browser Automation
A repeat-after-me training system for AI agents. Record your task, AI parses the video & DOM input, then replay your Doppel on demand.
The Problem
Automating repetitive desktop and browser tasks typically requires programming knowledge. Non-technical users need a way to teach AI by demonstration, not code.
My Role & Approach
Served as PM on a three-person team. Defined the product vision, designed the user flow, and coordinated between two developers during a hackathon sprint.
Tech Stack
Result
Won 3rd place at AI Hackathon, December 2023. Demonstrated the viability of demonstration-based agent training.
.png)
.png)
Cyber Fort
AI-Simulated Cyber Attacks for LLM Security
Leverages synthetic user knowledge to simulate attack vectors and build a self-improving firewall for AI companies.
The Problem
LLM applications are vulnerable to prompt injection, jailbreaking, and data extraction attacks. Static security rules can't keep pace with evolving threats.
My Role & Approach
Applied learnings from Spellforge's synthetic user framework to simulate adversarial attack patterns. Designed a self-improving defense loop that adapts to new attack vectors.
Tech Stack
Status
Hackathon project, January 2024
.png)
.png)
Connexpert
Digital Marketplace for Expert Interviews
A two-sided marketplace connecting subject matter experts with researchers and consultants for paid interviews.
The Problem
Researchers spend excessive time sourcing and vetting domain experts. Experts lack efficient platforms to monetize their knowledge for short consultations.
My Role & Approach
Hired as Head of Product (CEO-level). Led market research, defined the MVP, managed the build with developers, and executed go-to-market strategy.
Tech Stack
Status
Production, October 2022
.png)
.png)
OLEO
Dietary Supplement eCommerce Brand
Co-founded a DTC supplement brand. Raised $2.5M in funding and grew revenue to nearly $1M annually.
The Story
Co-founded in 2016 and built from the ground up: product development, branding, supply chain, and a custom-built eCommerce store. Raised a seed round and scaled operations.
My Role
Co-Founder. Oversaw product development, eCommerce operations, fundraising, and brand strategy. Managed the full lifecycle from concept to ~$1M ARR.
Key Metrics
Status
Series Seed, founded 2016
.png)