SmartJoules - Software Engineer - Python/Machine Learning

About the role :

As a Software Engineer on the Intelligence team, you will work across the systems that turn raw sensor readings into decisions : data pipelines processing telemetry from thousands of pieces of equipment, machine learning models that predict how that equipment behaves, and an AI agent that answers operators' questions from live building data.

This is a deliberately broad role. In your first year you will touch data engineering, applied machine learning and backend work, and you will find out which of them you enjoy most. You will work alongside senior engineers who review your code and explain their reasoning, and with product managers and customers who will tell you directly whether what you built is useful.

We do not expect you to know anything about buildings or HVAC when you start. Most of us did not.

Tech Stack :

• Languages : Python (primary), JavaScript, TypeScript

• Machine learning : scikit-learn, XGBoost, LightGBM, Keras, pandas, numpy, AWS SageMaker

• Reinforcement learning and control : Model Free RL, Deep RL, ML based Optimiser

• AI systems : AWS Bedrock, tool-calling agents, retrieval-augmented generation

• Data platform : Apache Airflow, InfluxDB, PostgreSQL, Redis, S3

• Knowledge graph : Amazon Neptune, Brick schema

• Observability : Grafana, Prometheus, Loki, Sentry, OpenTelemetry

• Backend and front end : FastAPI, Node.js, Angular

• Edge : ARM Linux, MQTT

• Cloud and infrastructure : AWS, Docker, GitHub Actions

What you will do :

• Explore and clean real sensor data - the kind with gaps, drift, stuck values and units that disagree - and work out what it is actually telling you.

• Build and maintain data pipelines that turn raw equipment telemetry into the metrics our models and our product depend on.

• Train, evaluate and retrain machine learning models on plant sensor data - power draw, water flow, temperatures, and the efficiency baselines used to verify savings.

• Write tools for our AI agent : functions that fetch data from our knowledge graph and time-series stores and return it in a form a language model can reason over.

• Trace data problems across the stack, from a sensor on a machine through message transport, storage and APIs to the dashboard someone is looking at.

• Write tests, particularly for the cases that have already caught us out once.

• Demonstrate your work - to the team regularly, and to customers as you grow into it.