SN Saam Nazempour

Cambridge MPhil DIS | ML/AI Engineer

ML engineer and quantitative developer building forecasting, inference, and applied AI systems.

Physics-trained data scientist with project work in Bayesian modelling, radiomics explainability, LLM time-series forecasting, and applied AI workflows in financial services.

Email GitHub Download CV
Quant research Quant development ML engineering Data science RAG evaluation

Selected work

Applied modelling work with clear technical outputs

LLMs for time series Python, LoRA, HuggingFace

LoRA-tuned LLMs for nonlinear forecasting

Fine-tuned Qwen2.5-0.5B-Instruct with LoRA and LLMTIME-style tokenisation to forecast nonlinear Lotka-Volterra systems under a strict FLOPs budget.

LoRA model trained forecast plot LoRA rank and learning rate sweep plot Context length training loss plot
  • Median R2 above 0.85 across 10 evaluation systems.
  • Tracked training and inference FLOPs, keeping total compute below 1e17.
  • Evaluated with MSE, correlation, DTW, and long-horizon forecast plots.
View repository
Bayesian inference JAX, NumPyro, HMC

Antikythera mechanism hole reconstruction

Built an inference pipeline estimating the original calendar ring geometry from 81 surviving holes, comparing isotropic and radial-tangential noise models.

Antikythera posterior predictive geometry plot
  • Used MLE, HMC/NUTS, posterior predictive checks, and residual analysis.
  • RT model estimated 355.26 +/- 1.41 holes with stronger WAIC performance.
View repository
Radiomics ML SHAP, sklearn, XGBoost

Radiomics feature selection and explainability

Developed end-to-end imaging biomarker workflows for photoacoustic and CT data, combining ANOVA sensitivity analysis, feature selection, classification, and model explanations.

ANOVA feature sensitivity plot SHAP beeswarm feature importance plot Classifier performance comparison plot
  • Compared RFE and Boruta workflows across original and resampled datasets.
  • Repeated SHAP analysis over 1000 seeds to assess explanation stability.
View repository
Scientific computing CASA, Python

Self-calibration for interferometric imaging

Built and evaluated a CASA-compatible self-calibration pipeline for VLA L-band observations of 3C147 and Abell S1063.

Radio interferometric imaging calibration plot
  • Used iterative gain correction to improve image fidelity.
  • Logged RMS noise and PSNR across calibration cycles.
View repository

Experience

Experience across applied AI, quantitative analytics, and scientific software

Lloyds Banking Group

Data Science & AI Graduate, ML/AI Engineer | Sep 2025 - Present

  • Developing Kubeflow components and orchestration pipelines for a GenAI RAG solution.
  • Evaluating chunking strategies and retrieval algorithms to improve context relevance.
  • Building RAG evaluation pipelines with Ragas, internal libraries, AI-as-judge workflows, and human review.
  • Developing internal user interfaces with HTML and Plotly Dash.

UBS

Quant Summer Analyst | Jun 2023 - Aug 2023

  • Built a dashboard for ACQA quantitative comparative analytics used by traders.
  • Designed models to visualise Greeks and sensitivity to market movements.
  • Created an iterative solver for barrier levels and rebates on custom structured products.
  • Back-tested a 10-year spread strategy and delivered findings into a monthly sales pack.

Education

Strong mathematical and computational base

University of Cambridge

MPhil, Data Intensive Science | Distinction, 76%

Machine Learning and AI, Deep Learning, Research Computing, Statistical Data Analysis, Bayesian Statistics, Software Development.

Imperial College London

MSci, Physics | 2:1

Computational Physics, Quantum Field Theory, General Relativity, Quantum Optics, Mathematical Methods, Differential Equations.

Technical stack

Tools used across modelling, ML, and deployment work

Python SQL GCP CI/CD NLP Kubeflow Git Bash Jupyter JAX NumPyro scikit-learn SHAP Plotly Dash Pydantic

Open to roles

Open to quant, quant dev, ML engineering, and data science opportunities.

Get in touch GitHub profile