Mariam Hassan

AI Researcher ยท DeepMind Scholar

Personal Website

AI Safety and AI for science with a focus on reliable biomedical systems.

AI researcher and DeepMind Scholar with an MSc in Artificial Intelligence for Biomedicine and Healthcare (Distinction) from UCL. I build research systems spanning spatial transcriptomics, low-resource NLP, and reproducible evaluation tooling.

Email LinkedIn GitHub Download CV

Research Experience

Graduate Research Thesis, UCL / The Alan Turing Institute

Jan 2025 - Oct 2025

  • Identified biological plausibility gaps in spatial transcriptomics generation from histology on HEST-1K breast cancer data.
  • Developed a biologically constrained LUPI framework using matched bulk RNA-seq during training while preserving H and E-only inference.
  • Built a multi-objective validation pipeline covering SSIM, GSVA pathway correlation, Moran's I, cell-type deconvolution, and gene-category analysis.

NLP Research, Low-Resource Languages, UCL

Jan 2025 - Apr 2025

  • Compared static and dynamic tokenization for LLaMA-3.1-8B and Gemma-7B on morphologically complex languages.
  • Implemented embedding-aware vocabulary extensions that reduced Arabic OOV rates by 17.4%.
  • Developed benchmarking for fertility, compression, inference efficiency, and downstream tasks with LoRA fine-tuning.

Undergraduate Research Thesis, Cairo University

Sep 2022 - Sep 2023

  • Developed a hybrid metaheuristic for dynamic vehicle routing with uncertain demand and real-time map changes.
  • Combined fuzzy receding horizon control with genetic algorithms and shipped a demonstrator with Flutter, Python, and Firebase.

Selected Research Projects

Biologically-Constrained Spatial Transcriptomics Framework

Research system using LUPI and multi-objective constraints for biologically consistent transcriptomics generation from histopathology.

Low-Resource Tokenization Benchmarking

Embedding-aware tokenization extensions and evaluation pipelines for morphologically rich languages with measurable OOV reductions.

Dynamic Vehicle Routing Research System

Hybrid optimization under changing conditions combining fuzzy control, genetic search, and real-time assignment policies.

Professional Experience

Cloud AI and Data Analytics Consultant, Bizbrain

May 2021 - Aug 2024

  • Led end-to-end client engagements across long-running consultancy projects.
  • Designed and deployed cloud AI solutions on Azure and AWS, reducing compute costs by 42%.
  • Built scalable MLOps and data pipelines that improved processing speed by 2.5x.
  • Implemented analytics and reporting workflows that reduced manual effort by over 250 hours per month.

Skills

Research and Modeling

Transformers, multimodal learning, diffusion models, reinforcement learning, GANs, Bayesian methods, tokenization and cross-lingual NLP.

Frameworks and Tooling

PyTorch, TensorFlow, JAX, Hugging Face (Transformers, Tokenizers, Datasets), scikit-learn, NumPy, Pandas, SciPy.

Cloud and MLOps

Azure ML, AWS SageMaker, Docker, Kubernetes, MLflow, Weights and Biases, DVC, CI/CD.

Programming

Python, C++, C, Java, R, Julia, MATLAB, SQL, Bash.

Education

MSc in Artificial Intelligence for Biomedicine and Healthcare, UCL

Sep 2024 - Oct 2025

Distinction. DeepMind Scholar.

BSc in Computers and Artificial Intelligence, Cairo University

Sep 2019 - Sep 2023

Contact

Email: mariam.ihab.mo@gmail.com

LinkedIn: mariam-ihab-mohammed

GitHub: fcistud