Anuvind Pramod

Anuvind Pramod

BS-MS Graduate, IISER Pune · Master’s Thesis, Jug Lab, Human Technopole, Milan

BS-MS (Interdisciplinary) · Graduated May 2026

I recently graduated from IISER Pune with a BS-MS (Interdisciplinary) degree, completing my Master’s thesis at the Jug Lab, Human Technopole, Milan, advised by Dr. Florian Jug. My thesis developed deep learning models for biological age estimation from screening mammography, using the “mammographic age gap” as a novel imaging biomarker for personalized breast cancer risk stratification.

My interests span medical image analysis, causal inference, self-supervised and transfer learning, and interpretable AI for clinical decision support — problems where computation can directly influence patient outcomes.

Seeking PhD positions in AI for medicine, causal inference & clinical AI · Available Fall/Winter 2026

Research

Biological Age Estimation from Mammography for Breast Cancer Risk Stratification code ↗

Developed an end-to-end deep learning framework for predicting biological age from screening mammograms — achieving a MAE of 3.13 years, demonstrating competitive performance against existing benchmarks. The architecture combines distilled DINOv3 backbones with hierarchical multi-head attention. Investigated mammographic age gap (biological − chronological age) as a novel imaging biomarker for personalized cancer risk assessment. Trained on large-scale multi-institutional datasets using HPC / SLURM infrastructure.

Deep CNN for Histopathology Image Classification code ↗

Designed a custom CNN with stacked convolutional blocks, batch normalization, and progressive dropout for multi-class lung and colon cancer classification. Achieved 99.58% test accuracy and AUC 0.9999 on the LC25000 benchmark. Conducted ablation studies comparing architectural variants against ResNet baselines.

Single-Cell Transcriptomic Analysis of YAP/TAZ in Breast Cancer

Developed a reproducible Python/Scanpy scRNA-seq pipeline for the lab. Investigated co-expression and functional redundancy of YAP/TAZ transcriptional co-activators in breast cancer vs. normal mammary epithelial cells using dimensionality reduction, clustering, and differential expression analysis.

Spatial Protein Expression Analysis in Breast Tumor Microenvironment

Analyzed THBS1 expression across tumor cell types using multiplex immunofluorescence imaging (PanCK and CD3 markers). Performed quantitative spatial analysis using Halo and inForm software.

Computational Optimization for Sustainable Aviation Fuel Production

Led the Measurements division (statistical modeling via Design of Experiments) and the Wet Lab team. Outcome: Best Education Award, nominated for Best Climate Crisis Project, Top-10 global undergraduate ranking at iGEM Jamboree, Paris.

Publications & Talks

Technical Skills

Machine Learning & AI

PyTorch scikit-learn ViT / DINOv3 / Swin CNNs Attention Mechanisms Transfer Learning Fine-tuning GradCAM SHAP Weights & Biases TensorBoard

Programming & Computing

Python C++ Bash HPC / SLURM Distributed Training Git Linux Docker uv

Medical Image Analysis

Mammography Histopathology Multiplex IF Halo / inForm DICOM preprocessing Whole-slide imaging

Bioinformatics & Data Science

scRNA-seq Scanpy Seurat Statistical Modeling Design of Experiments Matplotlib Seaborn AutoDock

Generative AI & LLMs

PEFT / LoRA QLoRA NLP Embeddings Tokenization AI Safety & Guardrails RAG Systems Vector DBs

Relevant Coursework

Machine Learning Natural Language Processing Deep Learning Computer Vision Statistical Inference Linear Algebra Bayesian Approaches Computational Biology Cancer Biology Molecular Cell Biology Systems and Implementation of Algorithms

Honors & Awards

Leadership & Service

Blog

Understanding the Causal Transformer (Melnychuk et al., ICML 2022)

How do transformers reason about counterfactual outcomes under time-varying treatments? This post unpacks the Causal Transformer — a method for estimating individual treatment effects over time using attention mechanisms adapted for causal inference, enabling more credible predictions of what would have happened under alternative treatment sequences.

Read more →