Ten years, from robotics labs to cell therapy.

Technical lead for single-cell multi-omics in oncology and immunology, from discovery through product development.

• Lead single-cell multi-omics programs (transcriptomics, proteomics, spatial) that feed target selection, mechanism-of-action work, and translational strategy for oncology and immunology assets.
• Set the analytical direction across several parallel projects: defining standards, reviewing methodology, and agreeing deliverables with clinical and discovery stakeholders.
• Mentor junior scientists and manage external consultants, mostly through code review, design discussions, and onboarding.
• Work with wet-lab, clinical, and CMC teams to turn analyses into experimental designs and go/no-go decisions.

Built the team's production analysis infrastructure and led single-cell projects across clinical and manufacturing programs.

• Built automated Nextflow/Python pipelines for scRNA-seq, CITE-seq, and multi-modal analyses, replacing manual workflows that previously took days per dataset.
• Led the single-cell analyses behind biomarker discovery for a clinical-stage cell therapy and process characterisation for CMC, including work that went into regulatory documentation.
• Set up the team's reproducibility conventions — containerisation, version-controlled pipelines, shared QC standards — now the default for new projects.
• Reviewed colleagues' analyses as the team's main technical reviewer.

PhD on open-source methods for spatial transcriptomics, now used by labs in academia and industry.

• Designed and led Optocoder, a machine-learning pipeline for decoding barcoded transcripts from imaging-based spatial transcriptomics. Published in NAR Genomics & Bioinformatics and used by external labs.
• Rewrote novoSpaRc (optimal-transport reconstruction of tissue architecture from scRNA-seq) for scale, making it usable on whole-tissue datasets. Co-authored the Nature Protocols paper.
• Co-inventor on a US patent for 3D spatial gene-expression reconstruction.
• Handled the full cycle for both tools: problem framing, method development, benchmarking, release, documentation, and user support.

Short collaboration applying protein language models to agricultural biotech.

• Ran an independent short-term project applying protein language models (ProtBERT) to find candidate anti-insecticidal proteins from raw sequence data, combining transfer learning with biological priors to prioritise candidates for validation.

Deep-learning system for pharmaceutical quality control.

• Developed deep-learning models for automated particle detection and classification from flow-microscopy images used in biopharmaceutical QC.
• Packaged the whole thing — ingestion, training, evaluation, reporting — as a Python/PyTorch/TensorFlow tool that internal scientists could run themselves.

Multi-omics integration for activity-dependent neuroscience.

• Applied supervised ML to activity-dependent changes in the neuronal proteome.
• Built a multimodal integration and domain-adaptation pipeline linking proteomic and transcriptomic datasets — the cross-modality work I still rely on today.

Real-time perception for an image-guided biopsy robot.

• Developed real-time needle-tip localisation and tracking from ultrasound imaging for autonomous robotic control during biopsy.
• Wrote the C++/CUDA interface connecting real-time image analysis to robot control.