StableMarch 20, 2026
AI Network System
5G-slicing decision engine for high-density wifi — predicts congestion per zone with scikit-learn and re-allocates bandwidth in real time.Yashwanth Kamireddi · ML & Frontend Lead
AI Network System is a decision-support platform for high-density wireless environments — colleges, stadiums, public events. It predicts congestion per zone and re-allocates bandwidth across high / medium / low priority 5G slices in real time, before users feel the latency spike.
A scikit-learn classifier — trained on per-zone signal strength, bandwidth usage, latency, packet loss, user count, and time-of-day — labels each zone LOW / MEDIUM / HIGH congestion in real time. A 5G-slicing engine then reallocates bandwidth across priority tiers using the prediction. The frontend dashboard visualizes everything live: zone status on a Leaflet map, congestion forecasts on Recharts time-series, and slicing decisions as they happen.
This was a multi-team build. I owned the React + Vite + Tailwind frontend end-to-end (Dashboard, Monitor, Predict, Slicing pages — Leaflet zone maps, Recharts time-series, real-time WebSocket-style updates) and the scikit-learn ML predictor (feature engineering, model training, joblib persistence, heuristic fallback so the demo stays live if the model fails to load).
- Backend — FastAPI + Uvicorn, scikit-learn model.pkl, REST surface (/simulate, /predict, /allocate, /health)
- Frontend — React 19 + Vite + Tailwind, React Router 7, Leaflet for zone maps, Recharts for time-series, Lucide icons
- Resilience — heuristic fallback when the ML model is unavailable; OpenAPI docs at /docs for the entire surface
