AI / MLApr 2026
EV Charging Intelligence
ML demand prediction + agentic trip planner

The problem
EV drivers don't know whether a charging station will be free when they arrive. A question like “I'm driving to New York on Friday at 5 PM with a Level 2 charger — where and when should I charge?” needs both a prediction and reasoning.
How it works
- 01User question
- 02Reasoner (Llama 3.3)
- 03ML model + live context
- 04Personalised itinerary
What I built
- Data pipeline for cleaning, time-series feature engineering and de-duplication.
- scikit-learn Pipeline (ColumnTransformer + RandomForestRegressor) for sub-second utilisation predictions.
- Four-node LangGraph state machine: Reasoner → ML tool + live-context tool → Synthesiser.
- Streamlit app with an analytics dashboard, interactive map and chat-based trip planner.
Highlights
ML as an agent tool
The trained model is called as a tool inside the agent, so answers rest on predictions rather than guesses.
Transparent reasoning
A typed state object collects every node's output, so you can see how each recommendation was made.
Lite model for deploy
The full 1.47 GB model is swapped for a <100 MB version so it runs on Streamlit Cloud.
Stack
PythonLangGraphGroqscikit-learnStreamlit
Next case study
HPCL Lead Intelligence Agent