Aarav Mahajan

AaravMahajan

5th Year Statistics and Data Science @ UBC

prev. Data Platform Engineer Co-op @ Sanofi

ExperienceSelected projects

Experience

Sanofi

Data Platform Engineer (Co-op)

Sep 2025 – Aug 2026

Toronto, Canada

  • Built an evaluation pipeline for a RAG data-search platform with W&B Weave, Amazon Bedrock, and Snowflake Cortex. Traced 19 agent methods with 4 scorers and automated regression tests in GitHub Actions.
  • Automated Snowflake deployments for a Streamlit app serving 3,000+ monthly users. Cut data-publishing time from 30+ days to 7.
  • Upgraded 40+ Lambda functions and 25+ Airflow pipelines to Python 3.13. Rebuilt EMR Serverless deployments with Docker, ECR, Terraform, and GitHub Actions.

Undergraduate Research Opportunities

Research Experience Committee, UBC

Aug 2024 – Apr 2025

Vancouver, Canada

  • Built an internal matching tool using the Gale-Shapley algorithm to pair 400+ undergraduates with 90+ graduate research mentors.
  • Recruited graduate students and postdoctoral fellows as research mentors. Increased mentor participation by 40%.

University of British Columbia

Research Assistant

Sep 2023 – Apr 2024

Vancouver, Canada

  • Built physics-informed neural networks in PyTorch to solve inverse differential-equation problems. Optimized physics-based loss functions to reach around 0.1% error against analytical solutions.
  • Ran 80+ numerical experiments across varying boundary conditions to test model robustness.

Selected projects

AWS Infra Automation

Scoped AWS roles and repeatable Lambda builds and deployments.

  • Terraform
  • AWS IAM
  • GitHub Actions
  • Lambda
  • Docker
Terraform Bootstrap (opens in a new tab)Lambda pipeline (opens in a new tab)
  • Managed project IAM roles with Terraform and S3-backed state. GitHub Actions authenticates through OIDC and applies a saved plan.
  • Built Lambda function and layer artifacts with Docker. Terraform manages the function, role, EventBridge schedule, and CloudWatch logging.
  • Terraform
  • GitHub Actions
  • Lambda
  • Docker
From change to infrastructure
  1. AuthenticateGitHub OIDC · temporary AWS credentials
  2. PlanGenerate and save the Terraform plan
  3. ApplyApply that same plan to project IAM roles
S3-backed state. Scoped permissions. One saved plan.

Physics-informed neural networks

Neural networks for inverse differential-equation problems.

  • PyTorch
  • Built physics-informed neural networks in PyTorch. Incorporated governing equations into the loss function.
  • Compared results with analytical solutions. Varied boundary conditions to investigate robustness.
  • PyTorch

URL Shortener

A Kubernetes service with deployment and observability at its core.

  • Kubernetes
  • Flask
  • Prometheus
  • Grafana
View repository (opens in a new tab)
  • Deployed a Flask/Gunicorn URL shortener on Kubernetes. Configured ingress routing and TLS certificates.
  • Tracked latency, traffic, errors, and saturation with Prometheus and Grafana. Added structured application logs and alerting.
  • Kubernetes
  • Flask
  • Prometheus
  • Grafana

Retex

Ask retail-data questions in plain language and get SQL-backed results.

  • LLaMA 3
  • FastAPI
  • PostgreSQL
  • React
View repository (opens in a new tab)
  • Connected React to a FastAPI backend. Used LLaMA 3 for model-assisted SQL generation.
  • Prepared retail data with a Python ETL workflow. Stored it in PostgreSQL for querying.
  • FastAPI
  • PostgreSQL
  • React

Billboard Longevity Prediction

A reproducible workflow for predicting album chart longevity.

  • R
  • Docker
  • GNU Make
  • Quarto
View repository (opens in a new tab)
  • Modeled Billboard chart longevity using Rolling Stone’s Top 500 albums. Cross-validated k-nearest-neighbors regression.
  • Connected data preparation, validation, training, and Quarto reports with GNU Make. Used Docker for a shared computational environment.
  • R
  • Docker
  • Quarto

Airbnb Revenue Prediction

Predicting rental revenue through feature engineering and model comparison.

  • R
  • glmnet
  • caret
View repository (opens in a new tab)
  • Imputed missing values and encoded categorical features. Compared XGBoost, Ridge, and Lasso regression.
  • Tuned models with cross-validation. Selected Lasso using mean absolute error.
  • R

Telehealth Access Disparities

Exploring demographic and geographic patterns in telehealth adoption.

  • SQL
  • Oracle Database
  • Python
View repository (opens in a new tab)
  • Cleaned public telehealth data into relational tables. Joined them in SQL to assemble analysis datasets.
  • Used hypothesis tests to examine adoption differences. Interpreted associations without claiming causality.
  • Oracle Database
  • Python
More work on GitHub (opens in a new tab)

A little
about me.

I’m interested in how complex systems behave, and what it takes to make them reliable.

I study Statistics and Data Science at UBC. What I've come to enjoy most isn't just finding solutions to problems, but understanding why those solutions work, what assumptions they rely on, and where they fall apart.

That curiosity has taken me beyond the mathematics. I spent a year at Sanofi working on data platforms and cloud infrastructure, building and maintaining the systems that move and process data. It gave me a different perspective on computing, one where reliability, architecture, and the details of implementation matter just as much as the ideas behind them.

These days, I'm particularly interested in machine learning, distributed systems, and the engineering that makes modern AI possible. As I figure out exactly where I want to go with those interests, I enjoy the process of taking something apart intellectually, understanding it properly, and putting that understanding to use.

Get in touch.

Always happy to talk about interesting problems, ideas, or opportunities.

LinkedIn (opens in a new tab)GitHub (opens in a new tab)