Ground

Schemas & Validation: Built JSON/Protobuf schemas for sensor readings with validation scripts, folder-level checks, and CI integration (GitHub Actions + Spark cleaning job).

Graph-Based Background Removal: Implemented classical segmentation for foreground masks and cut-outs, with CLI/API wrapper and post-processing.

Disease Anomaly & Worsening Detection: Developed offline pipeline for detecting anomalies and worsening trends in disease data using statistical baselines, Z-score/IQR/CUSUM, alert deduplication, and Postgres logging.

Soil Moisture Detection Pipeline: Built real-time CV pipeline (FastAPI + MobileNetV3) for wet/dry classification with threshold-based alerts, Kafka messaging, DLQ retries, and Postgres event logging.

End-to-End Integration: Connected full pipeline from MinIO image ingestion → Kafka → Flink processing → inference service → DB updates → PyQt GUI visualization of zones, history, and irrigation status.

PyQt GUI: Created interactive map showing all sprinklers, active zones, last images, history, and parameter updates

End-to-End Leaf Disease Detection Pipeline A comprehensive pipeline that ingests raw leaf imagery, performs automated preprocessing, applies YOLO-based leaf detection, extracts individual patches, classifies disease symptoms, generates structured metadata, and stores all outputs for real-time analytics and downstream processing.

Kafka Stream + Embedding Service Workflow A streaming workflow in which a Kafka consumer ingests imagery-notification events, validates and normalizes them, batches requests to a CLIP-based gRPC embedding service for vector generation, enriches each message with embeddings, and publishes the enhanced events to downstream Kafka topics with DLQ handling.

End-to-End Leaf Disease Service (MinIO → OpenCV/Model → PostgreSQL): Built an end-to-end Python service that pulls leaf images from MinIO, runs an OpenCV disease detector, and writes structured results into PostgreSQL, turning raw images into traceable leaf-disease reports in the AgCloud pipeline.

Leaf Disease Detection Logic (Multi-stage ML Training): Built a three-stage ResNet18 training pipeline (PlantVillage → PlantDoc fine-tuning) using PyTorch, Albumentations and MixUp, to classify each leaf as healthy or sick and assign a specific disease class with robust performance on real-field images.

Leaf Disease Dashboard (Desktop + Grafana Integration): Developed a PyQt6 “LeafDiseaseView” dashboard that queries leaf reports from PostgreSQL via a REST API, computes key KPIs, ranks devices and diseases, and embeds a Grafana drill-down view per disease and date range for interactive field-level analytics.

Kafka + Flink Automated Test Lab with PyTest & Testcontainers: Implemented an automated Kafka+Flink “test lab” using PyTest and Testcontainers that spins up ephemeral clusters, streams hundreds of test images through the pipeline, and verifies exactly-once processing with high branch coverage as part of the CI pipeline.