Artifacts

Human-in-the-loop AI system built for NYU SPS admissions. Handles routine follow-up, missing-document reminders, and post-admission task tracking while keeping staff in control of judgment calls. Built with a 5-agent pipeline with confidence-threshold routing, eliminating manual handling for routine cases while ensuring every AI decision is traceable and reviewable.

Comparative sentiment analysis of customer reviews using a Generative AI (LLM-based) model vs. a traditional ML pipeline in RapidMiner. GenAI outperformed on sarcasm and mixed-sentiment language; traditional ML won on auditability. Findings translated into a brand perception report for stakeholders.

Automates the daily market research that PMs do manually. Monitors Reddit, Hacker News, Stack Overflow, and the web for competitor signals. Uses Claude to synthesize findings into a concise email digest. Runs on a daily schedule or on demand, with a web UI for managing subscriptions. Scales to zero on AWS.

AI-powered IT self-service portal built on AWS Bedrock and Claude. Auto-grants or escalates software access requests based on job-level policy. LLM chat interface lets employees find tools in natural language and managers look up staff context.

End-to-end statistical analysis of U.S. honey production data, examining supply-side dynamics including production volumes, pricing trends, and inventory levels across states and time periods. Applies inferential statistics and exploratory data analysis to surface actionable insights for agricultural commodity forecasting.