Machine Learning Intern
📅 Posted 1h ago
📄 Job Description
About Us
LANE is building an AI-based driving intelligence platform that detects and scores real-world driving behavior from mobile sensor data. We’re a small, deep-tech team solving hard signal-processing and pattern-recognition problems with real vehicle data — not toy datasets.
About the Role
This is a full-time internship opportunity for 3–6 months, based in Bengaluru.
What You'll Do
- Analyze time-series sensor data from real driving sessions to identify behavioral patterns and events
- Design, train, and validate ML models for event classification and severity scoring
- Own the full model lifecycle — data prep, feature engineering, training, evaluation, and iteration
- Drive empirical threshold calibration on real, noisy session data over theoretical assumptions
- Debug and improve detection accuracy against ground-truth annotated data
Requirements
Must-Have Skills
- Strong ML fundamentals — you deeply understand how models actually learn: loss functions, gradient descent, bias-variance, regularization, overfitting, train/val/test discipline, evaluation metrics
- Hands-on model-building experience — you’ve personally implemented and trained models on real projects (not just run notebooks end-to-end). Be ready to walk us through one in depth
- Strong mathematical fundamentals — statistics, probability, linear algebra, hypothesis testing
- Python for ML — NumPy, Pandas, scikit-learn; PyTorch or TensorFlow
- Analytical rigor — comfort forming and testing hypotheses against messy, imperfect real-world data
Good to Have
- Time-series / signal processing exposure (filtering, windowing, frequency analysis)
- Edge/mobile ML deployment (TensorFlow Lite / ONNX)
- Prior work with IMU, GPS, or embedded sensor data
- Git and collaborative workflows (Jira)
Who You Are
A builder, not a tutorial-follower. You’ve trained models yourself, debugged why they didn’t work, and can explain every design choice you made. You’d rather dig into messy real-world data than tune a pre-built model on Kaggle — and you’re comfortable saying “the data doesn’t support that assumption” and backing it up.
What You'll Gain
- Direct ownership of models that ship into a live consumer product
- Mentorship from engineers with automotive, embedded, and applied ML backgrounds
- End-to-end ML exposure: from raw sensor data to production detection logic
