Data Scientist – Multimodal AI & Computer Vision

About the Project
We are developing an animal-welfare monitoring platform designed to build an individual understanding of animals over time.
Starting in an operational cat shelter in Greece, the platform will combine video, audio, sensor data, human observations and medical records to help identify meaningful changes in behaviour, health and welfare.
This is a genuinely early-stage project. This is a genuinely early-stage project. Camera infrastructure is already in place, but the data and AI platform is being built from the ground up. The technical direction is intentionally open: the solution may use conventional computer-vision models, vision-language models, multimodal models, or a combination of these approaches.
The Role
We are looking for a hands-on Data Scientist to help design, build and evaluate the first AI capabilities of the platform.
Your work will focus on turning raw, imperfect real-world data into practical experiments and working implementations. You will help determine which technical approaches are appropriate, rather than simply implementing a predefined architecture.
A central challenge will be moving beyond the analysis of isolated camera events. The platform should progressively build a longitudinal profile of each animal by combining observations across time, locations and data sources.
You will work alongside an experienced Senior Research Data Scientist, who will contribute research depth and help shape the technical direction, and a Systems Engineer responsible for the physical camera, sensor and infrastructure environment.
This is a practical development role. It is highly focused on building, testing and learning quickly rather than producing theoretical research.
We are open to a strong junior or mid-level candidate. Depth of practical ability, curiosity and learning speed matter more to us than years of experience or job title.
What You'll Do
Explore and work directly with real-world video, audio, sensor, human observations and other shelter data.
Build practical prototypes and baselines to test different technical approaches.
Develop and evaluate computer-vision capabilities where appropriate, potentially including detection, segmentation, tracking, re-identification, pose estimation and behaviour or action recognition.
Experiment with vision-language models and multimodal approaches for video understanding, information extraction and behavioural interpretation..
Evaluate when specialised models, foundation models or hybrid approaches are most appropriate.
Develop methods for linking observations to individual animals across cameras, locations and time.
Explore how different data sources can be combined into an evolving profile of each animal.
Develop individual baselines and approaches for detecting meaningful behavioural or physiological changes.
Train, fine-tune, evaluate and compare models.
Reproduce and adapt promising approaches from current research.
Build reproducible training, evaluation and inference pipelines.
Work with GPU-based environments and optimise video and model processing where required.
Work closely with the Senior Research Data Scientist to translate research ideas into practical experiments and working implementations.
Collaborate with the Systems Engineer as new cameras, sensors and data sources are introduced or modified
Document experiments, results and technical decisions so that we can learn and iterate quickly.
What We're Looking ForEssential
Strong Python programming skills and the ability to produce clear, maintainable code.
Practical experience building, training and evaluating machine-learning or deep-learning models.
Experience with PyTorch, TensorFlow or equivalent.
Hands-on experience working with image, video or other high-dimensional unstructured data.
Practical exposure to one or more relevant modelling tasks, for example object detection, tracking, re-identification, segmentation, pose estimation, action or behaviour recognition, or multimodal modelling.
Ability to take an idea or research approach and turn it into working code.
Sound understanding of model training, validation and performance evaluation.
Ability to structure datasets and create repeatable preprocessing, training and evaluation workflows.
Comfortable working with incomplete, noisy and evolving real-world data.
A pragmatic and experimental mindset, with the ability to learn quickly and change direction based on evidence.
We do not expect candidates to have prior experience across every method listed above.
Particularly Interesting
We would be especially interested in candidates with experience or strong academic/project exposure to one or more of:
Vision-language models or multimodal LLMs
Video understanding
Audio, time series or sensor data modelling
Anomaly detection, change detection or individual baseline development.
Edge AI, NVIDIA GPU environments or efficient model inference.
Building early-stage ML products or research prototypes from limited data.
Animal behaviour, animal monitoring or other real-world observational systems.
What We Don't Require
You do not need a PhD or a long list of publications.
You also don't need to have already worked in animal welfare.
We are open to someone relatively early in their career if they have strong technical foundations, have genuinely built things themselves and are excited by the opportunity to learn quickly alongside an experienced researcher.
Working Environment
This is not an established ML team with mature datasets, production pipelines and predetermined specifications.
You'll be joining while the system and technical direction are still under development. That means exploring data, testing approaches that may not work, making pragmatic decisions with incomplete information and helping determine what should be built next.
For the right person, there is significant scope to grow with the project and take increasing ownership as the platform develops.

Work arrangement
Yes