Projects

My vision was a true end-to-end B2C app: traffic acquisition (social, Google, Meta ads) → working signup flow → subscription billing → full analytics for insights and experimentation. Proud to say I accomplished this, slowly and surely. I went through several cycles of getting 90% there, only to find myself stuck in AI doom loops. Through prompt engineering trial and error, experimenting with different coding LLMs, and plenty of refactoring, I finally realized the vision. I'm no longer developing databot as I pursue more targeted (practical) projects. Key learnings: AI coding blasts open new possibilities to make me a better DS/ML practitioner, by contributing to the entire development lifecycle. Don't be afraid to start over. And solve specific problems first—generalize within that constraint.

My current passion project! I play in 2 home leagues—a 14-team Mixed dynasty in its 19th year where I'm the defending champ 😊🏆, and a 30-team Mixed Franchise league with 20-man minor league rosters. I also compete in multiple NFBC leagues where a few thousand crazies vie for infamy and a $200k grand prize. The goal: productionize the hacky scripts and predictive models that have driven my success into a real application. At the core is the Dataset Wizard—a master dim_player table that recognizes players from almost any datasource and enforces templates for high-quality prediction modeling. Layer in your league metadata (standings, rosters, player values, schedules, trades, FA pickups, history) and you get analytics views unavailable anywhere else: projections customized to your league, all your leagues in one spot, and AI-powered recommendations.

After wrapping up databot, I realized chatbots alone aren't unique—you need specific use cases. I accompanied a friend (a rainwater harvesting specialist) on a graywater management job, took some photos, and within 2 hours had a working site built on the askdatabot framework. Just from pictures, it generated Statements of Work, customized quotes/invoices, and hour-by-hour task lists—the recommendations were incredible. Sizing and product pricing needed work, but the insight was clear: with receipts and an audit trail of actual work performed, this could become an entire contractor backoffice system. I left it as a proof of concept, but it gave me a newfound understanding of how to use AI appropriately—solving specific problems for real customers.