Telemetry Data Scientist

Telemetry Data Scientist / Real-Time ML Engineer Division:DATUM, Impac Exploration Services Location:Remote or Oklahoma City Type:Full-Time or Contract Build Models That Run at 10,000 Feet Per Second We're not looking for someone to build another dashboard. We're looking for someone who sees streaming sensor data and thinks "I could predict that failure 30 seconds out." DATUM is where industrial operations meet cutting-edge ML. We're streaming terabytes of WITS telemetry from drilling operations—pressure, torque, vibration, flow rates—all live from assets worth millions per day. This isn't historical analysis. This is real-time ML where your models make decisions while the drill bit is turning. Here's what makes this interesting: The physics are unforgiving. The data is messy. The stakes are real. And when your anomaly detection prevents a catastrophic failure at 10,000 feet? That's the kind of win that gets noticed—by operators, by industry, and yeah, by recruiters from places you've dreamed about working. What You'll Actually Build • Streaming ML pipelines that process millions of sensor readings per minute • Anomaly detection that understands the difference between "weird but fine" and "stop everything now" • Predictive models that see patterns in vibration harmonics humans miss • Real-time feature engineering on streams that never stop • Decision systems that earn trust from people who've been reading these signals since before you were born Technical Reality Check Core stack: • Streaming:Kafka/Pulsar, Flink/Spark Streaming, or River (bonus if you've pushed these to their limits) • ML Ops:MLflow, Weights & Biases, or Kubeflow—because models in notebooks don't save rigs • Time Series:Prophet, LSTM/GRU implementations, or custom architectures for sequential data • Languages:Python (NumPy, Pandas, Scikit-learn) + SQL + whatever else gets the job done • Deployment:Docker, Kubernetes, edge computing frameworks • Monitoring:Grafana, Prometheus, or custom solutions—because you can't fix what you can't see Nice to have: • Physics-informed neural networks or domain-aware ML • Experience with industrial protocols (OPC-UA, MQTT, Modbus) • Edge computing / model optimization for embedded systems • That one project where you made TensorFlow run somewhere it shouldn't You're Our Person If • You've built something real-time that stayed up when it mattered • Sensor fusion excites you more than computer vision • You can explain Kalman filters and gradient boosting to a drilling engineer • You've debugged model drift at 3 AM because production doesn't sleep • "It works in my notebook" isn't in your vocabulary Even Better If • You've touched industrial systems, IoT, or anything with real sensors • Built telemetry systems for fun (racing sims, drone projects, even elaborate Raspberry Pi setups) • Contributed to time-series or streaming ML libraries • You understand that 99.9% uptime means 8.76 hours of downtime per year—and that's too much The Culture Small team. Big problems. No bureaucracy. We move fast, ship faster, and celebrate when our work is so good that other companies try to poach you. You'll have the compute you need, the freedom to experiment, and colleagues who push you to be better. Fair warning: This isn't a "set it and forget it" role. When a $100M operation depends on your model, you'll feel it. But when your algorithm catches something nobody else saw? That feeling is addictive. Why This Matters While everyone else fights over chatbot improvements, we're building ML for the physical world. Energy is complex, consequential, and desperately needs fresh thinking. Come help us prove that the best telemetry systems don't just live at race tracks—they're 10,000 feet underground, running 24/7, keeping people safe and operations efficient. Ready to build ML that matters? Let's talk. We are not sponsoring visas or participating in CPT programs at this time.