Data scientist and machine learning engineer, Grenoble, France
I work across data science and deep learning: LLM agents that plan a task, write the Python for it and run it; vision models adapted with few labels; and pipelines over images, signals and sequencing data. The through line is making moderate-size models useful through careful adaptation and honest evaluation rather than through scale. Clinical data is where I have gone deepest, because it punishes anything that only works on a benchmark.
01 About
My work splits between two things that are usually kept apart: building with generative models, and making predictive models dependable. On the generative side I build agents that take a stated objective, reason about how to reach it, write and execute the Python that does the work, and then check their own result. I do the same kind of work professionally on the agentic capabilities of a large language model product used by more than 500 million people.
On the predictive side I spend my time on the parts that decide whether a model is useful in practice: adapting it with few labels, building the data and annotation workflows around it, and designing evaluation that survives contact with real cases. I have done this on MRI, whole-slide histopathology, ECG and sequencing data, and on vision-language models, where I tested where a general model stops being dependable for localisation.
02 Experience
Turing, remote, independent contractor
Optimising the agentic capabilities of a large language model product with a user base above 500 million, and earlier, structuring large-scale datasets to establish its performance metrics.
GIPSA-Lab (CNRS), Inria Grenoble Rhone-Alpes and CHU Grenoble Alpes
Master's thesis on automatic quantification of skeletal muscle in pediatric MRI. Measured how far SegmentAnyMuscle, a foundation model pretrained on adult MRI, transfers to children, and characterised its failure modes along acquisition, anatomy and pathology on a clinical cohort of nine children.
Khyber Medical University, Peshawar
Deep learning on gigapixel whole-slide histopathology to reduce subjectivity in cancer grading, under a grant co-funded by the Higher Education Commission Pakistan and the British Council UK. In parallel, GPU-accelerated sequencing pipelines with NVIDIA Clara Parabricks to align, call and annotate variants and prioritise candidate oral-cancer genes.
AI in Healthcare Lab, National Center of Artificial Intelligence, Peshawar
Classification of atrial fibrillation, bradycardia and tachycardia from ECG, and sound source separation and denoising over spectrograms with generative adversarial networks.
03 Publications
Exploring the mutational spectrum of key kinase genes PIK3CA, BRAF, EGFR, ALK and ROS1 in oral squamous cell carcinoma
doi:10.1186/s12885-025-14609-8Profiling genetic mutations in the DNA damage repair genes of oral squamous cell carcinoma patients from Pakistan
doi:10.1038/s41598-025-91700-xSingle-channel speech enhancement using colored spectrograms
doi:10.1016/j.csl.2024.10162604 Projects
Generative AI and agents
A Streamlit application in which Gemini 2.5 Pro turns a stated preprocessing objective into executable Python: it inspects each image, writes and runs the processing code locally, visualises the result and judges whether the objective was met. Conversations branch from any earlier message, so alternative pipelines can be compared without losing the history. Code at github.com/fazeel15/SpectraWeaver.
Digital pathology
Detection and classification of squamous cells in histopathology images with object-detection networks, producing case-level grades backed by evidence a pathologist can inspect rather than a single opaque score.
Physiological signals
ECG preprocessing, handling of class imbalance, and evaluation of multi-class cardiac arrhythmia models with metrics suited to the imbalance.
Genomics
Alignment, variant calling and annotation of raw FASTQ data with NVIDIA Clara Parabricks, then downstream analysis and visualisation to prioritise candidate genes.
05 Skills
06 Contact
For data science and applied AI roles, research collaboration and PhD positions, or anything about the work above.
khanfazeel15@gmail.com