Harshil Suthar

AI-Based Smart Traffic Light Controller

A physical 20×30" prototype intersection — 12 streetlights, 4 traffic lights, one Raspberry Pi Pico W — running a closed loop between a YOLOv8 vision model and a Random Forest decision model trained on SUMO simulation data.

My role: embedded development in MicroPython, rigorous hardware testing, and the serial protocol between the CPU vision pipeline and the Pico actuators.

Camera (Pixel 10 + DroidCam)
  → YOLOv8 vehicle detection
  → Random Forest decision model
  → Serial (USB) → Pi Pico W
  → LED traffic + street lights

AI-Driven Crop Recommendation System

Built in 8 hours of live coding under blind-dataset evaluation. A Random Forest regressor predicts crop yields from soil composition (pH, N, P, K), weather forecasts, and rotation history, then ranks recommendations on a dual axis — yield vs. sustainability (water and fertilizer efficiency).

Why it scored well: the pipeline was built fault-tolerant against unknown dataset formats — judges threw a blind dataset at us on presentation day and it handled it without modification.

EcoEats — Photo to Recipe

Full-stack app that takes a photo of your fridge, identifies ingredients, and generates recipes from what you already have. Built to cut household food waste.

Photo → Azure Vision AI → ingredient list
      → Azure OpenAI    → recipe generation
      → Flask backend   → Tailwind frontend