I build machine learning systems that turn raw data into decisions.

An AI-ready web interface for real-time surveillance of irrigation canals and water courses, built to replace manual patrolling with automated digital oversight and cut water theft and distribution inequities.

Real-time computer vision pipeline reaching 92% detection accuracy by fine-tuning YOLOv8 on custom training data and integrating DeepSORT for multi-object tracking, with outputs stored in MongoDB for downstream analysis.

Content-based filtering engine using TF-IDF feature vectors over a normalized, indexed PostgreSQL schema — cut recommendation latency by 30%.

End-to-end ETL pipeline ingesting 500K+ records from heterogeneous sources into a PostgreSQL warehouse, cutting report generation time by 45% through schema normalization and query optimization.

Pricing-trend models across 80K+ transactions with 87% directional accuracy, backed by SQL queries and stored procedures powering interactive dashboards for stakeholders.

Multi-agent chatbot architecture with sub-agents, grounding responses in live database queries via structured retrieval — improved factual accuracy by 38%. In progress

Lets a user drop in a YouTube URL and ask questions about the video — pulls the transcript via the YouTube API and answers using retrieval-augmented generation.