Dr. Amara Osei
Co-founder & Chief Scientist
PhD, Machine Learning — University of Cambridge. Formerly deep RL research at DeepMind. 40+ peer-reviewed publications.
About & Research
A research lab that ships
CogniPulse Labs was founded on a simple frustration: brilliant research rarely survives deployment, and deployed systems rarely benefit from real research. We closed the gap — a PhD-led team that publishes at top venues and ships models into regulated, high-stakes production every quarter.
- Founded in London
- 2019
- Researchers & engineers
- 38
- Hold a PhD
- 22
- Peer-reviewed publications
- 90+
Leadership
Led by researchers, staffed by engineers
Every practice area is led by a PhD with deep domain research experience — and every engagement pairs them with production engineers who have run systems at planetary scale.
Dr. Lukas Brenner
Co-founder & Head of Engineering
PhD, Distributed Systems — ETH Zürich. Built inference infrastructure serving 2B+ requests/day at scale.
Dr. Mei-Lin Chen
Director of Computer Vision
PhD, Vision & Robotics — Imperial College London. Led industrial inspection research at Fraunhofer IOSB.
Dr. Ravi Chandran
Head of Predictive Analytics
PhD, Statistical Learning — University of Oxford. Former principal forecaster at a FTSE-10 energy group.
Dr. Sofia Marchetti
Head of AI Ethics & Governance
PhD, Philosophy of Technology — TU Delft. Advises two national regulators on deployed-AI assurance.
Dr. Tomas Halvorsen
Head of Edge AI
PhD, Embedded Computing — NTNU. Holds 9 patents in on-device model compression and acceleration.
Research
Selected academic citations
Our team publishes at KDD, NeurIPS, FAccT, MLSys, and IEEE venues. A selection of recent work directly informing our client deployments:
Osei, A., Brenner, L., et al. (2025). "Calibrated Confidence for Streaming Fraud Ensembles under Distribution Shift." Proceedings of KDD 2025, Applied Data Science Track.
Chen, M.-L., Halvorsen, T. (2024). "Line-Speed Wafer Inspection with Distilled Vision Transformers on Industrial Edge NPUs." IEEE Transactions on Industrial Informatics, 20(11).
Marchetti, S., Osei, A. (2024). "Operationalising Model Cards: Auditable Governance for Deployed Enterprise LLMs." FAccT 2024, Industry Track.
Chandran, R., et al. (2023). "Hierarchical Probabilistic Forecasting for Renewable Grid Balancing with Calibrated Intervals." NeurIPS 2023 Workshop on ML for Physical Sciences.
Brenner, L., Halvorsen, T. (2023). "Shadow-Mode Rollout: Safe Fleet-Wide Model Updates for On-Device Inference." MLSys 2023.
Governance
Our AI ethics framework
Ethics is not a committee that meets quarterly — it is a set of engineering constraints enforced in every evaluation harness, deployment review, and client contract.
Measured claims only
Every performance number we publish is measured in production against a client-agreed baseline. If a metric cannot be measured, we do not claim it.
Human accountability by design
High-stakes decisions retain a human in the loop. Our systems augment and flag — they do not silently decide about people.
Bias & fairness testing
Domain, demographic, and drift testing runs in every evaluation harness before deployment, and continuously after. Results are shared with clients in full.
Data minimisation & privacy
We train on the least data that works, prefer on-premise and VPC-isolated inference, and never reuse client data across engagements. Ever.
Refusal to build
We decline projects we believe will cause harm — surveillance of individuals, manipulative systems, or applications we cannot make auditable. This is a standing policy, not a case-by-case judgement.
Transparent failure modes
Every deployment ships with documented failure modes, fallback behaviour, and an ops runbook. A model whose failure modes are unknown is not deployed.
Request the full framework document
The complete CogniPulse AI Ethics & Governance Framework (v3.2) is available to prospective clients under NDA.