Anthony Kovari
Projects
ML Capstone — United Airlines Aircraft QA
Four-model ML pipeline (garbage filter, interior/exterior classifier, paint quality scorer, sentiment analyzer) surfaced in a React reviewer dashboard. ETL ingests images from email and social media.
Full-stack production ML platform monitoring land-use change across 13,000 km² of the Chicago metro using real Sentinel-2 satellite imagery (2019–2026). ResNet-18 fine-tuned on EuroSAT achieves 97–99% per-class accuracy and exports to ONNX. Imagery is ingested from the Copernicus CDSE API; models are versioned in MLflow; FastAPI serves predictions on AWS EKS with full Terraform infra-as-code, GitHub Actions CI/CD, and Prometheus/Grafana observability.
Computer Vision — Satellite Image Classification
End-to-end ML system classifying satellite imagery by land use. ResNet-18 trained on EuroSAT; CLAHE preprocessing via OpenCV; real-time inference through a containerized FastAPI web UI.
Edge AI — Track Invasive and Endangered Birds
Solar-powered edge device running BirdNet on a Google Coral Edge TPU to autonomously identify bird species in the wild. Sightings are logged in real time to a citizen science database.
Agentic AI — LLM Workflow Automation
LLM-orchestrated agentic pipeline that coordinates parallel reasoning tasks across a structured multi-step workflow. An automated data collection layer (custom scrapers + APIs) feeds a FastAPI backend that fans out to parallel LLM calls, streams results incrementally to the UI via SSE, and surfaces them in a human-in-the-loop review dashboard for inspection and action.