PhD Track AI Research

Mechanistic Interpretability &
Geometric Deep Learning

Investigating how neural networks generalize high-dimensional data into geometric manifolds to learn causal patterns in complex biological systems.

"My research generalizes beyond standard bioinformatics. I focus on the geometry of high-dimensional data, using mechanistic interpretability to understand how latent variables drive system behavior—building AI that is not just predictive, but architecturally transparent and secure."

Data -> Latent Manifold -> Logic

Professional Journey

Nov 2018 CSIR (Secure AI) PierianDx UVA (Causal Research) Present

Technical Stack & Theory

AI Research & Theory

Mechanistic Interpretability Geometric Deep Learning Manifold Learning Latent Variable Discovery Bayesian Networks

High-Dim Data

Dimensionality Reduction Graph Neural Networks (GNN) 80k+ Variable Modeling Hypothesis Testing Complex Systems

Engineering & Security

Privacy-Preserving AI Split-Key Architecture Docker / AWS PyTorch / TensorFlow MLOps Pipeline

Key Projects: Modeling & Architecture

Causal Discovery in High-Dim Space

UVA | BIMS PhD Track (Computational Biology)

Developed a Bayesian Network framework to disentangle causal drivers from noisy, high-dimensional datasets (80k+ variables). Focus on mapping complex dependency graphs to identify latent "driver nodes."

Causal Inference Graph Theory

RAPID-CT: Decoupled AI

CSIR | Lead Architect

Engineered a privacy-preserving inference architecture. Deployed a ResNet-based model where the input manifold (Image) and Identity (PII) were processed in decoupled, encrypted streams.

Secure Inference Computer Vision

Systems Genetics & Manifolds

Farber Lab | UVA

Applied Systems Genetics to model complex traits. Used statistical learning to map genotype-to-phenotype relationships, effectively treating biological variance as a high-dimensional optimization problem.

QTL Mapping Statistical Learning

Unstructured Data Mining

UVA | PhD Track

Processed heavy unstructured text/string data (long-read sequencing) to resolve novel signals. Enhances data completeness for downstream predictive modeling.

NLP / String Processing Feature Engineering

Selected Publications

J. Am. Soc. Mass Spectrom.

Multi-Omic Pipeline Validation

Integrating unstructured long-read sequencing data with targeted mass spectrometry. A case study in multi-modal data integration.

PMC / PubMed

Complex System Regulation

Investigating novel regulators in biological barriers. Demonstrates application of hypothesis testing in complex physiological systems.

Education

University of Virginia

PhD Candidate, Computational Biology

Current (Exp 2027) | Focus: Applied High-Scale ML Systems

KIIT, India

B.Tech & M.Tech (Dual Degree)

Biotechnology (Computational Focus)

Saikat Bandyopadhyay

Mechanistic Interpretability & Geometric Deep Learning