Senior Machine Learning Engineer

About Bytebeam

Bytebeam builds the hardware and software stack for next-generation connected vehicles and IoT devices. Our platform helps teams collect telemetry, monitor device health, debug issues remotely, push OTA updates, generate alerts, and operate connected fleets at scale. We are now building the intelligence layer on top of this data: algorithms that convert raw vehicle signals into diagnostics, predictions, recommendations, and real-world business outcomes.

Role Overview

We are looking for a Senior Machine Learning Engineer to build production-grade algorithms for connected vehicles, EVs, and fleet systems.

This is not a dashboard-only analytics role. You will work with noisy field telemetry, CAN/OBD/J1939 signals, fault codes, battery and powertrain data, sensor behavior, and service outcomes. You will design algorithms that detect issues, predict failures, explain root causes, recommend action, and continuously improve as more vehicles come online.

You should be comfortable moving from messy data exploration to deployable models, from signal-level debugging to fleet-level insights, and from interesting graph to customer can take action now.

What You'll Do

Build vehicle intelligence algorithms

Develop algorithms across areas such as fuel, EV battery health, diagnostics, driver behavior, and predictive maintenance. Example problem areas include:

• Fuel theft, refill, drain, consumption, fill-rate detection, and fuel-sensor failure detection.
• Fuel rail pressure, pumping efficiency, oil pressure, turbo-health, air-filter, coolant-temperature, and powertrain anomaly detection.
• DEF efficiency, DPF health, air-intake-system performance, and after-treatment diagnostics.
• Ambient-aware EV range prediction, battery SOH/SOC estimation, battery-pack temperature forecasting, and battery-swap readiness signals.
• Driver-behavior scoring and fuel-efficiency impact attribution.
• Fault-code aggregation, diagnostic-code sequencing, RCA, upstream failure analysis, and guided repair recommendations.

Own the full algorithm lifecycle

Take algorithms from data discovery to production. This includes data cleaning, signal validation, feature engineering, model development, offline evaluation, field validation, deployment, monitoring, drift detection, and iteration.

Work with real-world vehicle telemetry

Handle missing data, sensor noise, inconsistent calibration, vehicle variants, firmware differences, network gaps, and edge cases from field deployments.

Partner across teams

Work closely with firmware, hardware, validation, cloud, product, customer success, and field teams. You will use bench data, vehicle logs, service feedback, and customer reports to make algorithms reliable outside the lab.

Build explainable and actionable outputs

Translate model outputs into alerts, diagnostics, probable causes, recommended fixes, confidence levels, and fleet-level insights. The goal is not just prediction. The goal is useful action.

Must-Have Skills

• 4+ years of experience in applied ML, data science, algorithm engineering, signal processing, or time-series analytics.
• Strong Python skills, including NumPy, pandas, scikit-learn, and production-quality data pipelines.
• Experience with time-series data, anomaly detection, forecasting, classification, regression, rule-based systems, or hybrid ML plus heuristics.
• Strong SQL and comfort working with large telemetry datasets.
• Experience converting messy sensor data into robust features and measurable algorithms.
• Ability to evaluate algorithms using precision, recall, false-positive rate, latency, confidence, coverage, and field impact.
• Experience productionizing models or algorithms, not just building notebooks.
• Strong debugging instincts and ability to reason from first principles.

Strongly Preferred

• Automotive, EV, IoT, fleet, industrial, or embedded-systems data experience.
• Familiarity with CAN, OBD-II, J1939, UDS, DTCs, ECUs, telematics devices, or vehicle logs.
• Experience with EV battery systems, BMS data, SOC/SOH estimation, thermal models, or range prediction.
• Experience with diesel/powertrain systems such as DPF, DEF, turbo, fuel rail, oil pressure, coolant temperature, and air intake.
• Experience with streaming data, Kafka/Flink/Spark, Airflow, data lakes, or real-time analytics.
• Exposure to online learning, reinforcement learning, auto-calibration, Bayesian methods, Kalman filters, survival models, causal inference, graph-based diagnostics, or sequence models.
• Ability to build simple internal tools, dashboards, notebooks, or scripts for validation and debugging.

Good-to-Have

• Experience deploying algorithms to edge devices or constrained environments.
• Experience working with firmware and hardware teams.
• Knowledge of MLOps, model monitoring, experiment tracking, and model governance.
• Rust, Go, or systems-programming exposure.
• Prior work in predictive maintenance, remote diagnostics, fleet intelligence, insurance telematics, energy analytics, or connected vehicle platforms.

What Success Looks Like

First 30 days: Understand Bytebeam's telemetry schema, vehicle signal sources, customer use cases, and current alerting/diagnostics stack. Identify the first 2 to 3 high-value algorithms to build or improve.

First 90 days: Ship one validated algorithm into pilot use, with clear metrics, alert definitions, dashboards, and field feedback loops.

First 6 months: Own a small portfolio of production algorithms across vehicle health, diagnostics, fuel/energy, or EV battery intelligence. Reduce false positives, improve diagnostic accuracy, and create reusable algorithm frameworks for future use cases.

Ideal Candidate

You are a builder who can speak both ML and machines. You enjoy data, but you do not hide inside notebooks. You can sit with a field log, a firmware engineer, a DTC sequence, and a customer complaint, then emerge with an algorithm that actually helps someone fix a vehicle faster.

You care about precision, but also about practicality. You know that the best algorithm is not always the fanciest one. Sometimes it is a calibrated rule, sometimes a state machine, sometimes a tree model, sometimes a forecast.