WO/HBO - Bachelor or Master ᐧ AI / ML / Computer Science or similar ᐧ 5–10 months starting in August 2026 or later ᐧ Hybrid (remote/office) ᐧ € 400 - € 700 / month Who we are Praevi is a startup founded by two experienced entrepreneurs. It's born from the daily frustrations that people in healthcare have with the existing work rostering processes. Poor work-rostering drives up sick-leave and staff turnover while at the same time the demand is rising due to an aging population. Existing solutions don't solve this, so we started Praevi to target this from a different angle. Vincent van den Tol (technical co-founder, computer science background) is responsible for the product and technology. Léon Janssen (commercial co-founder) leads the relationships with healthcare organizations and commercial growth. We're now looking for an AI/ML or computer science graduate to join our growing team and improve the personal alignment and flexiblity of automated planning processes. Why you should consider this assignment Let's face it, everybody want to do something with AI and this is probably not the only assignment you will read. But while most assignments are focused on saving costs, identifying risks or finding new business opportunities, we offer the possibility to have a direct impact on society by helping healthcare organizations, nurses and physicians to keep healthcare accessible. Context Dutch healthcare is under strain: demand is growing while staffing falls short. Rostering ("who works when") is critical to using hours efficiently and keep people happy and energized. Instead, it's often a major source of frustration due to inflexibility and lack of autonomy. Many optimization algorithms exist, but we see that a lot of healthcare teams don't use them, or keep a manual process on the side. We believe this is because of: Formal mathematical solvers require correct and complete problem modelling. That is hard and time-consuming to do, especially in a domain as dynamic as healthcare. Complex systems often act as a "black box" for people, and therefor lack the feeling of control, trust and transparency. As a result, qualified staff spend days per month on manual planning, changes require constant back-and-forth, louder voices win, and rules depend entirely on the individual planner — hurting job satisfaction and retention. Our approach Rather than optimizing existing solutions by a few percent, we believe we can have more impact by first making automatic rostering accessible to everyone in healthcare. This requires us to think differently: Problem modelling We should be able to easily translate your current (offline) planning process into an automated planning model Dynamic rosters Applying changes or testing new scenarios to existing rosters should be done fast . Transparency Solutions should be transparent and explainable ("why wasn't she assigned to that shift?") Personallized approach A solution should learn people's (personal) preferences to produce better-aligned rosters We see an opportunity to combine LLMs with proven ML algorithms, modern interfaces, and agent-based execution to implement these requirements. Our vision is a system that takes varied inputs (files, conversations, assignments, new facts) and turns them into a problem model for an automatic solver. A solver that surfaces hidden requirements, learns rules and preferences, can be run over and over, updates and improves dynamically, and keeps users in control.* Our application is already running inside a first hospital in The Netherlands and we are in the process of setting up pilots with more organizations. We have set the groundwork, and your work will contribute to expanding this foundation into our long-term vision of a smart, flexible and user-friendly planning solution.

The assignment Our central question is the following: How can we use AI agents to translate human input into a dynamic problem model for a mathematical solver such as MILP, and make sure the result and decisions made by the solver are explainable and transparent to staff? When we can automate the problem modelling step with an AI-agent that understands human language, can interpret existing (excel) files and document, and can discuss with people to help find hidden rules or explain trade-offs, it will become possible for any planner or team-leader to automate their planning proces. We're open to reformulating the question based on new insights, as long as it fits our long-term vision. Two references give initial direction: An LLM-powered MILP modelling engine for workforce scheduling guided by expert knowledge — https://arxiv.org/abs/2511.02364 The Rhythm of the Roster (MILP optimization for Dutch healthcare) — https://essay.utwente.nl/essays/108598 For your assignment you will work in a small but experienced team. You will have freedom to work how and where you like, you can (optionally) participate in related developments if that helps your assignment, and work with very direct lines and regular face-to-face moments. To research and test your solutions we will provide you with technical tools & infrastructure and connect you with healthcare providers in our network. Technology As a privacy-focused company in a sensitive sector, we focus on open-source frameworks and 100% European (cloud) infrastructure. Your research and solutions will need to work within a Python based environment, using mainly open source technologies and frameworks. You will get access to AI tools for coding assistent and analysis. Practical details Level: [WO / HBO — Bachelor or Master] Background: AI / ML / Computer Science or similar Required skills: Python; affinity with optimization/MILP and LLMs is a plus Language: English or Dutch (Dutch is preferred for communication with healthcare professionals) Duration & time frame: 5-10 months, starting in August 2026 or later Compensation: € 400 - € 700 / month Location & work mode: Hybrid (our office is in Haarlem, 10 min walk from Central station) Deliverables: Besides a thesis we expect a working and tested prototype, documentation, and recommendations for improvements Supervision: Day-to-day support and technical guidance from Vincent (technical co-founder) How to apply: Reach out to Vincent on vincent@praevi.nl or +31641298581 Website: https://praevi.nl/ *Technical solutions always support a human process, so we also help clients with organizational change. That's out of scope here but may yield useful inputs.