24/7 live-streamed ambient installation with signed audio, oscilloscope visualizer, and live AI commentary — all at the edge.
Outcome: Zero-ops, low-latency experience fully hosted on Cloudflare; extensible dashboard for “strange events.”
Live, interactive audience with the Pope — real-time chat, playful confession flows, and generative UI.
Outcome: Engaging, shareable experience; robust real-time pipeline.
A tongue-in-cheek, high-polish web experience celebrating “good old fashioned traditions” with generative media.
Outcome: Strong visual identity and delightful microcopy; solid performance on modern stacks.
A playful classification concept exploring cultural memes and lightweight interaction.
Outcome: Quick-to-grasp concept with viral potential; fast, minimal deployment.
PROBLEM: Manual IT ticket triage is a bottleneck, causing inconsistent prioritization and slow resolution times for users.
SOLUTION: Developed a Mistral-7B Large Language Model that transforms raw ticket text into actionable insights, requiring zero model training. Using prompt engineering, the system instantly classifies a ticket's category, assigns its priority and ETA, tags key information, and drafts a complete initial response in a single process.
PROBLEM: Manual identification of plant seedlings is slow and error-prone, making efficient, large-scale crop management difficult.
SOLUTION: Developed a deep learning model with high accuracy that automates classification, confirming the feasibility of real-time mobile identification and automated monitoring systems for precision agriculture.
PROBLEM: Customer churn is an ongoing concern for credit card issuers. Identifying card holders likely to close their accounts is critical to proactive retention efforts.
SOLUTION: Developed a high-performance ensemble model (XGBoost) that successfully flags 85% of true churners, enabling the bank to focus its retention efforts surgically and maximize ROI.
PROBLEM: High marketing spend and low engagement on personal loan campaigns makes it difficult to identify high-potential customers.
SOLUTION: Developed a Decision Tree model to forecast customer interest, identifying clear indicators like income and education to slash outreach costs and boost conversion rates.
PROBLEM: Operational bottlenecks and incorrectly identified key market drivers lead to inefficient driver staffing and unfocused marketing efforts.
SOLUTION: Developed a full exploratory data analysis in Python that delivered an outline for growth by illuminating key operational and marketing insights. The analysis pinpointed consistent patterns of demand, enabling data-driven staffing models to improve service quality.
Original writing on technology, culture, and strange machines. Start with What Is An Altar?