Tech internship: Agentic and Context-Aware Edge AI for Adaptive Appliance Monitoring

Develop an adaptive edge-AI system that learns electrical devices and their operating modes through sensor data, user interaction and Agentic AI. Are you a tech-savvy student interested in machine learning, Agentic AI, IoT and edge computing? In this internship, you will build on an existing appliance-recognition and anomaly-detection pipeline and investigate how it can evolve into an adaptive system that can learn new devices and operating modes with limited manual configuration. The work will contribute to the RESCUE Horizon project, where intelligent monitoring of devices and building systems can support the detection, prediction and mitigation of potentially hazardous situations. **About this internship** We have developed an appliance-recognition pipeline that can identify not only which electrical appliance is connected, but also what that appliance is doing. The system also includes an unsupervised anomaly-detection pipeline that can detect abnormal behaviour. The next challenge is to make this approach generic, adaptive and user-driven. Instead of defining all devices and operating modes beforehand, the goal is to investigate whether Crownstone sensing combined with Agentic AI and human feedback can create a system that learns continuously from its environment. For an arbitrary electrical device, the desired system should be able to: * Detect or propose the type of device that is connected. * Interact with the user to determine which operating modes are important. * Collect labelled examples of these operating modes through normal device usage. * Use user feedback as an indirect supervised-learning mechanism. * Train or update models that recognise the learned operating modes automatically. * Adapt when new devices, new operating modes or changing user requirements are introduced. You will work with real sensor data and an existing technical pipeline rather than starting from scratch. At the same time, there is significant freedom to investigate new machine-learning, Agentic AI and human-in-the-loop approaches. A central research question is therefore how to create an evolvable context-aware AI pipeline in which the definition of relevant operating modes does not have to remain fixed over the lifetime of the system. **What will you do and learn?** * Study and extend the existing Crownstone appliance-identification, operating-mode, and anomaly-detection pipelines. * Design an adaptive device and operating-mode learning architecture that can be applied to different appliances. * Investigate how an AI agent can interact with end users to determine which operating modes of a device are relevant. * Develop a human-in-the-loop learning workflow in which users can confirm or reject device classifications and operating-mode predictions. * Investigate how user feedback can be converted into labelled training data and used to incrementally improve the underlying machine-learning models. * Develop an application or user interface for live data annotation, allowing users to: add new devices, define new operating modes, record additional examples of existing operating modes, label live sensor data, correct model predictions, and remove devices or operating modes that are no longer relevant. * Integrate an LLM that can interact naturally with the user to collect contextual information, request labels and validate model predictions. * Investigate approaches such as active learning, incremental learning or continual learning to allow the system to evolve when new classes or behaviours are introduced. * Evaluate different ML and LLMl architectures for deployment on resource-constrained edge hardware. * Build and evaluate a working prototype using real Crownstone electrical measurements and realistic device operating cycles. **About you** You are studying computer science, artificial intelligence, data science, electrical engineering, embedded systems or a related field, preferably in your final year of your Master. Ideally, you have the following qualifications: * Good Python programming skills. * Experience with machine learning and data analysis. * An interest in LLMs, Agentic AI or human-in-the-loop machine learning. * An interest in time-series data, signal processing, IoT or embedded systems. * Strong analytical and problem-solving skills. * The ability to work independently on an open-ended R&D challenge. It’s a bonus if: * Have previously worked with LangChain/LangGraph or similar agent frameworks. * Have experience with ML frameworks (e.g., PyTorch, TensorFlow, scikit-learn). * Enjoy research-oriented projects and experimentation. **About Almende** Join Almende to work on innovations that make the world better and smarter through self-organization and technology. We research complex problems, together with an international network of partners, and generate ideas to solve them. We are a small, flat organisation with short communication lines. You will have room to work independently, test ideas and turn research into working prototypes. If you enjoy taking initiative, have many ideas, and wish to execute them, you’ll feel right at home. **We offer** * Real hands-on experience with Agentic AI, machine learning, IoT and edge AI: apply what you have learned in your studies to real-life situations * The opportunity to build on an existing high-performance appliance-monitoring pipeline. * Experience with real high-frequency electrical sensor data and physical devices. * A research challenge combining LLMs, adaptive machine learning and embedded systems. * The opportunity to contribute to technology developed within a European Horizon research project. * Guidance and mentorship from experienced engineers and researchers. * Freedom to investigate and prototype new technical approaches. * A peek into the international tech scene * Build a network with industry pros * A hybrid working environment in Rotterdam. Interested? Apply now! Job Type: Internship Pay: €400,00 per month Ability to commute/relocate: * 3013 Rotterdam: Reliably commute or planning to relocate before starting work (Required) Work authorization: * Netherlands (Required) Work Location: Hybrid remote in 3013 Rotterdam