Dhruv saini
Machine Learning Engineer specializing in EDA,ML,Dl Python
New to Contra
Dhruv is building their profile!
Halo CME Detection & Space Weather Prediction System
Title: ML-Based Early Warning System for Solar Storms (Smart India Hackathon Winner)
Problem: Geomagnetic storms and Coronal Mass Ejections (CMEs) threaten satellite infrastructure, but predicting their arrival and severity is a hard time-series forecasting problem with multi-scale temporal patterns.
My approach: I developed supervised time-series forecasting models (LSTM, GRU) trained on space-weather data, applying lag and rolling-window feature engineering to capture temporal dynamics across multiple scales. I also built a real-time ML-based alerting pipeline for risk assessment and forecasted key geomagnetic indices (Kp, Ap, Dst, Sunspot Number) 7 days ahead.
Tools used: Python, PyTorch/TensorFlow, LSTM, GRU, time-series feature engineering, MLflow
Result: Achieved a 7% improvement in CME/storm prediction accuracy and a 6% improvement in arrival-time prediction. Won Smart India Hackathon 2025 (Ministry of Education's Innovation Cell, Government of India), and the work was published as corresponding author in the Journal of Science, Computing and Engineering Research (JSCER), 2026.
1
25
EV Charging Diagnostic AI Assistant
Title: AI-Powered Diagnostic Assistant for EV Charging Infrastructure
Problem: Ecoplug Energy needed a way to diagnose EV charger faults in real-time and let users trigger remote actions (like initiating charging) via natural language — without needing a human support agent on every call.
My approach: I built a Generative AI–powered diagnostic assistant with intent-driven conversational workflows for real-time fault detection, plus a WhatsApp interface that translates natural-language commands into backend actions. The core was an async FastAPI decision-engine with priority-based routing (Diagnostic → Action → Intent → AI), combining regex extraction, fuzzy matching, and O(1) indexed lookup across 598+ diagnostic codes.
Tools used: FastAPI, Python, AWS (EC2, S3, Lambda), Terraform, MongoDB, Pydantic, async architecture
Result: Shipped a production-grade, scalable REST platform with confidence scoring and text normalization, deployed via Render Blueprints with a live GitHub Pages demo — giving users instant fault diagnosis and remote charging control through a conversational interface.
0
30