Sidharth Shah

Sidharth Shah

I build machine learning for molecules and genomes.

Biomedical engineering, computing option, at the University of Waterloo. This fall I’m doing bioinformatics research at The Centre for Applied Genomics at SickKids.

EmailGitHubLinkedInDevpost
A bacterial enzyme from the histone deacetylase family, drawn as a ribbon. Each active site holds an inhibitor (pink) beside a zinc ion. Drag to turn it.

Now

Graph neural networks for autism genetics

SickKids, The Centre for Applied Genomics

Building a heterogeneous graph-based transformer (HGT) to rank every human gene by its ASD risk, using de novo variants from SPARK, SSC, and MSSNG data cohorts (>70,000 people).

Projects

8 projects shown

Drug–ligand binding affinity predictor

Drug discovery

Predicts how strongly a drug binds its target protein. Protein sequences go through ESM-2, ligands become molecular graphs, and five GNN variants are tested against XGBoost and random forest baselines.

  • Trained on 51,784 protein–ligand pairs
  • Splitting by protein cut the generalization gap by 32%

PyTorch Geometric, ESM-2, RDKit, XGBoost

Read the write-upCode on GitHub

  • FragFlow

    Generative chemistry

    In progress

    A GFlowNet that builds drug-like molecules one fragment at a time, rewarded on drug-likeness, ease of synthesis and solubility.

    torchgfn, PyTorch Geometric, ZINC250K

    Code on GitHub

  • CardioGlasses

    Wearables

    MHacks 2026 track winner

    Glasses with heart and motion sensors that track heart metrics and raise alerts. Won the FinchNode HealthTech track at MHacks 2026. I built the layer that turns raw sensor signals into decisions.

    PPG, IMU, FinchNode API

    Code on GitHub

  • See-DR

    Medical imaging

    Accepted to CUCAI 2026

    Finds several kinds of retinal lesions at once to help detect diabetic retinopathy, using our HydraLA-Net architecture. Built with my Wat.AI team.

    PyTorch, Segmentation

    Read the paper

  • Similar-property protein generator

    Protein design

    ESM-2 suggests small edits to a protein that keep its properties.

    ESM-2, PyTorch

    Code on GitHub

  • Diabetes prediction network

    Health

    Predicts diabetes onset from health measures found in routine medical records.

    PyTorch

    Code on GitHub

  • Mental health support chatbot

    Health

    A support chatbot that looks up trusted material before it answers (retrieval-augmented generation), built on the OpenAI API.

    OpenAI API, RAG

    Code on GitHub

  • Internship scraper

    Automation

    Checks company career pages on a schedule and records new internship postings. Later extended with a Claude-powered scraping agent.

    AWS Lambda, EventBridge, DynamoDB, Playwright

    Code on GitHub

Experience

  1. Machine Learning Researcher

    The Centre for Applied Genomics, SickKids

    Scherer Lab, affiliated with Dept. of Molecular Genetics at U of T

    Fall 2026 to now

    • Improving a heterogeneous graph transformer for autism genetics that models genes, patients and genetic variants together
    • Integrating the SPARK autism cohort, roughly doubling the training data

  2. Machine Learning Researcher

    Wat.AI

    See-DR team

    2025 to 2026

    • Co-developed HydraLA-Net, a CNN that segments four kinds of retinal lesions at once
    • Co-authored a paper accepted to CUCAI 2026

  3. Software Engineer Intern

    LeapAP

    2026

    • Built web scrapers for an accounts-payable automation platform

  4. Software Engineer Intern

    Bell Canada

    2025

    • Built and maintained features on a mission-critical internal app, including OAuth 2.0 sign-in
    • Managed CI/CD pipelines and AWS deployments

  5. QA Engineer Intern

    autonomik.ai

    2024 to 2025

    • Tested ML pipelines and services in Python on GCP with CI/CD

Writing

Working on ML for biology or drug discovery? Let’s talk.

s78shah@uwaterloo.ca