Senior ML Systems Engineers – Vision ML / Visual Data Processing Data Platform
A privately backed AI technology company operating at the intersection of machine learning in the media industry.
You will own core ML infrastructure end to end, from data ingestion and curation through to distributed training and production inference, working with large-scale multimodal datasets (video, embeddings, metadata).
This is not a research role. The focus is on productionising models, building reliable platforms, and making ML systems fast and scalable in a real production environment.
The ideal profile is an ML engineer transitioning from research into platform ownership – someone who is product-minded and outcome-driven rather than tech-for-tech’s-sake. You should be comfortable bridging the gap between experimentation and production.
Key Responsibilities
- Build and evolve a data platform (LanceDB, DataFusion, SQL and vector search) for large-scale multimodal datasets
- Design ML pipelines for video indexing and processing (face detection, quality assessment, tracking)
- Improve training performance across single and multi-node setups using PyTorch and Ray
- Build evaluation and experimentation systems (Parquet/Iceberg) for model output analysis
- Own model versioning, lifecycle management, and promotion to production
- Optimise inference pipelines using Triton; build model ensembles and define request protocols
Requirements
- Proven ML engineering background with a focus on infrastructure and productionisation (not just model training)
- Strong Python skills, plus experience with a robust production language such as C++ or Java
- Solid understanding of data pipeline performance trade-offs: I/O, compute, batching, memory layout
- Hands-on PyTorch experience: training pipelines, data loading, preprocessing
- Practical distributed systems experience (Ray, DDP, or similar)
- Experience handling TB-scale or high-throughput data pipelines
- Familiarity with columnar formats: Arrow, Parquet, Iceberg
Nice to Have
- Exposure to video or visual media pipelines (FFmpeg, encoding, frame extraction)
- Vector search or embedding system experience
- Triton or production inference background
- React/frontend for internal tooling
Hybrid working.
