About Trackbeat

Trackbeat is a GTA-based industrial IoT sensor manufacturer, started in 2025. We build wireless vibration, uptime/downtime and condition-monitoring sensors, plus the platform that turns their data into early warnings before machines fail. We have pilots running with manufacturers and are building the AI layer that sets us apart.

- You will be our first ML hire. You'll own the models that detect anomalies, classify faults and predict remaining useful life from vibration and machine-state data — from raw accelerometer signals to production alerts on our dashboard and, over time, on the edge device itself. Real labelled failure data is scarce, so you'll also build the synthetic and physics-informed data that lets us train before failures happen.

What you'll do

- Build anomaly detection, fault classification (imbalance, misalignment, looseness, bearing wear) and remaining-useful-life models from tri-axial vibration, temperature and run-state data.
- Engineer signal features: FFT, envelope spectra, RMS, kurtosis, crest factor, order tracking, bearing fault frequencies.
- Create synthetic and physics-informed training data, and run lab tests on our fan and motor rigs to produce labelled faults.
- Turn models into a production pipeline: ingestion from gateways/MQTT, feature store, training, versioning, monitoring and retraining.
- Set alert thresholds that operators trust — keep false alarms low and explain why an alert fired.
- Compress lightweight models to run on edge hardware (ARM / iMX95 SoM, later the MCU)



to cut bandwidth and battery use.
- Work with hardware (sampling rate, sensor placement) and with customers in pilots to collect feedback and ground truth.
- Help define the v2 roadmap: AI features no competitor offers.

What we are looking for

- 3–6 years in ML or data science, with at least one model you took to production and maintained.
- Hands-on experience building ML for industrial IoT sensors — vibration analytics, condition monitoring or predictive maintenance (e.g. at a PdM startup, an industrial OEM, or a research lab in this area).
- Python, NumPy/SciPy, pandas, scikit-learn, and PyTorch or TensorFlow.
- Anomaly detection and learning with few or no labels (autoencoders, isolation forest, one-class methods, semi-supervised).
- Ships end to end: data pipelines, APIs (FastAPI), Docker, cloud deployment, model monitoring.
- Product thinking: asks what the customer needs delivered, attacks problems from different angles, doesn't wait for a perfect spec.
- Explicit written communication; comfortable in a small, fast startup.

Nice to have

- ISO 18436 vibration analyst certification (Cat II or higher); ISO 10816/20816 and bearing fault physics.
- Edge ML: TensorFlow Lite, TinyML, ONNX, quantization, embedded Linux.
- MQTT, time-series databases (TimescaleDB, InfluxDB), Postgres/Supabase.
- Synthetic data, simulation or physics-informed ML.
- Public datasets experience: CWRU, IMS/NASA bearing, PRONOSTIA, MIMII.
- LLMs for explaining alerts and generating maintenance recommendations.

📌 Machine Learning Engineer (Delhi)
🏢 Trackbeat
📍 Delhi