About enliteAI
enliteAI builds AI and optimization systems for critical infrastructure, with a focus on electricity networks. We work with network operators and industrial partners on problems where the decisions carry real consequence: how a distribution grid is operated, how much capacity it can safely carry, and how flexible resources at the grid edge are coordinated. Our work runs from applied research through to systems that operate on real measurement data. We are part of several EU Horizon research programmes — AI4REALNET, AI-EFFECT and INSIEME — and we maintain Maze, an open-source framework for simulation based reinforcement learning. Our team combines reinforcement learning, optimization, data engineering and power systems expertise, and sits deliberately between academic research and industrial deployment. We are based in Vienna. We are a small team and work like one: cross-disciplinary by default, low-ego, and more interested in whether something holds up than in who proposed it. We are looking for an ML Research Engineer to develop the methods behind our energy applications, and to carry them from the literature to something a network operator can rely on. Distribution grids are increasingly operated close to their technical limits, while many operators lack full visibility of their own low-voltage networks. That combination produces a class of problems well suited to machine learning: inferring system state from sparse and imperfect measurements, forecasting under uncertainty, evaluating network constraints fast enough to act on them, and coordinating flexible resources at the grid edge. Current projects include state estimation and dynamic operating envelopes, and the emphasis will move as new work comes in. What stays constant is the character of the problem. These are open questions at industrial scale rather than solved ones, the data is real measurement data with everything that implies, the physical constraints cannot be negotiated with, and the decisions have to be made in operational time. We build and deploy these systems ourselves, so the methods you design end up running against real data, and you own that transition. Roughly half the role is research — literature, method design, and experiments that establish whether something actually works. The other half is engineering it into a system that survives production. Both halves are the job. Alongside this you will contribute to Maze, extending it toward energy and infrastructure applications. That work is publicly visible and used beyond enliteAI. You would join a team that combines reinforcement learning and optimization, data and platform engineering, and power systems expertise. That mix is deliberate: the physical plausibility of a result is something you can check with a colleague rather than guess at, and the infrastructure your experiments and deployments run on is owned by the team rather than left to you. The domain itself is learnable and we will teach it — we would rather appoint an excellent ML engineer and provide the power systems knowledge than proceed the other way around. Expect three to six months to working competence. "Research" in the title describes the nature of the work rather than a seniority bar: this suits someone completing or recently completed a doctorate, as well as anyone who has built comparable research experience by another route.
Your tasks
The Role
Tasks
- Develop machine learning and reinforcement learning methods for grid operation problems: learned and hybrid state estimators, surrogate models for fast constraint evaluation, graph neural networks that exploit network topology, forecasting under uncertainty, and reinforcement learning for control within operational limits
- Review and assess the state of the art, judging which published approaches are implementable at industrial scale and which remain limited to small test cases
- Design and run experiments with meaningful baselines, ablations and realistic data regimes, validated against established simulation environments (pandapower, PyPSA, OpenDSS, Grid2Op)
- Contribute to core components of Maze and extend it toward energy and infrastructure applications
- Own the model-facing side of our data pipelines: domain transforms, unit and sign conventions, feature definitions, train/serve consistency, and the schema contracts that define valid input data
- Take prototypes through to MVP, and stay with them through that transition
- Contribute to deliverables and publications in our EU-funded research projects, working alongside energy experts, researchers and ML engineers
Requirements
- Fluent English with strong communication skills — you can explain a method, and honestly its limitations, to ML colleagues, to power systems engineers, and to an operator deciding whether to rely on it
- Deep, hands-on machine learning and reinforcement learning expertise with strong PyTorch skills, including a clear sense of when these methods are not the right choice
- Strong Python and sound software engineering practice — research code that only its author can run is not a sufficient outcome here
- You can read a paper and reproduce it, including resolving what the authors leave unstated
- Rigour in experimental design: meaningful baselines, controlled comparisons, and appropriate scepticism toward your own preliminary results
- Self-directed, you don't need a detailed roadmap to make progress
- Low-ego and collaborative, and genuinely interested in learning at the intersection of machine learning and power systems
- Willing to own the full development cycle, from literature review through to a deployed MVP
- A degree in computer science, energy informatics, electrical engineering, applied mathematics or a related field, or equivalent practical experience
- Valid work permit for Austria
It would be great if you None of this is required, and nobody has all of it — any one is a useful signal.
- Have power systems knowledge: power flow, state estimation, dynamic operating envelopes, or the operational context of distribution system operators
- Have used grid simulation frameworks such as pandapower, PyPSA, OpenDSS or Grid2Op
- Have worked in an adjacent method area — optimization (AC optimal power flow, convex relaxations), time series forecasting, or graph neural networks
- Have MLOps or data engineering experience: experiment tracking, model deployment, streaming data (Kafka, MQTT), or time series at scale
- Have work in public — publications at venues such as NeurIPS, ICLR, PSCC or IEEE PES, or contributions to a large open-source codebase
- Speak German, which helps when working directly with Austrian and German network operators
Benefits
- Research problems that are genuinely open, on infrastructure that matters
- Both a research and a product dimension: our EU Horizon projects provide structure, visibility and the opportunity to publish, and behind them stands a longer-term objective of establishing these capabilities as products
- Publicly visible open-source work through Maze
- A team spanning reinforcement learning, optimization, data and platform engineering, and power systems — with the domain expertise to sanity-check a result and the platform expertise to run it
- Our own compute cluster, on our own hardware in a Vienna datacenter: dedicated GPU capacity for experiments rather than a shared queue or a cloud budget to argue for
- Hybrid working: 2–3 days per week at our office in the herat of Vienna's 1st district, with minimal core hours
- Dedicated time and budget for R&D, conferences and professional development
- Choose your own hardware and ...