My background in engineering and research.

I'm Ruman Shaikh. I work across AI engineering, scientific machine learning, and computational biology, and I completed an MSc in Biomedical Engineering at Imperial College London with a focus on computational bioengineering.

At IBM Client Engineering, I developed AI systems with research teams and clients, building agents, retrieval workflows, fine-tuned language models, and the evaluation needed to understand their behaviour. I also worked on COBOL-to-Java modernisation and monitoring models on client data infrastructure, which connected the modelling work with the software and operational requirements of each engagement.

Before that, at Oracle Health, formerly Cerner, I developed population-health pipelines and analytics using EHR and laboratory data for millions of patients. My responsibilities included CMS reporting tests, pipeline performance, workflow automation, and production support, alongside co-leading an annual CMS update and training junior engineers.

These roles made me interested in how a system is evaluated as well as how it is built, because a working demonstration does not tell us how it will behave across different data or operating conditions. Long term, I want to continue developing AI and computational methods that are useful in science and healthcare.

My dissertation at Imperial tested whether inverse models trained only on synthetic microbial ecosystems could transfer to sparse, noisy real mouse data. I built a generalised Lotka-Volterra benchmark comparing classical regression, supervised inverse networks, PINNs, PI-DION, a quantum-classical variant, and graph and attention-based interaction decoders. The clearest result was that graph-based relational decoding improved interaction recovery and transferred best under biological perturbation, while the physics residual provided a smaller and more conditional gain.

More broadly, I'm interested in biological priors, neural scaling behaviour, reinforcement learning, and operator learning, particularly when limited data makes it important to understand which assumptions a model relies on.