whoami

Sanjay Subramanian

Backend + AI engineer @ Waterloo

Bachelor of Mathematics in AI, Minor in Computer Science · University of Waterloo · Expected April 2029

4 software roles · built a world model that learns to walk with ~172× fewer real tries

// experience

  1. Software Developer @ Macaca AI

    Aug 2026 – Present

    Remote

    • Designed async job pipelines on Redis Streams with retry, backoff, and dead-letter queues to isolate slow providers.
    • Extended a RAG pipeline that chunks and embeds uploaded documents into Qdrant for per-tenant knowledge retrieval.
  2. Software Engineering Intern @ U Plus

    Jan 2026 – Apr 2026

    Richmond Hill, ON

    • Rebuilt the payment flow in React/TypeScript, boosting checkout conversion by 11% and cutting completion time by 35%.
    • Right-sized EC2 instances and added request-driven auto-scaling, cutting monthly cloud costs by $3,800.
  3. Software Engineering Intern @ Mercury Inc

    May 2025 – Aug 2025

    Remote

    • Deferred non-critical scripts and lazy-loaded checkout images, improving LCP by 210ms and bounce rate by 12%.
    • Wrote 180+ Jest and React Testing Library tests for the checkout flow, raising coverage from 41% to 85%.
  4. Software Engineering Intern @ Terre Sky

    Jan 2025 – Apr 2025

    Remote

    • Wrote modular Terraform scripts to provision all AWS infrastructure, cutting staging setup from 1 hour to 5 minutes.
    • Engineered a Rust data ingestion service processing 5,000+ events/second at sub-10ms latency.

// projects

The simulated humanoid walking forward

World model: walking robot + robot brain

2026
  • From-scratch C++20 world model (own matrix maths and neural nets) teaches a simulated humanoid to walk with ~172× fewer real tries than PPO.
  • Camera-view redraw lifts π0.5 from 57% to 73% on moved-camera tasks; 4-bit π0.5 runs in 3.3 GB and scored no worse over 630 paired tries.
  • C++20
  • MuJoCo
  • Python
  • PyTorch
  • π0.5

read case study →

Offline Voice Assistant

Jan 2026
  • Event-driven pipeline with asyncio queues, achieving under 800ms end-to-end latency.
  • Runs fully offline with faster-whisper for speech-to-text and Llama 3.2 via Ollama — no API costs.
  • Python
  • faster-whisper
  • Ollama
  • asyncio

Personal Finance Tracker

Sep 2025
  • Full CRUD with AG Grid editable tables and real-time validation, cutting data entry errors by 25%.
  • JWT auth with bcrypt hashing for 100+ accounts with row-level data isolation.
  • React
  • TypeScript
  • FastAPI
  • PostgreSQL
  • Docker

This website

2026
  • Terminal-style portfolio with an AI chatbot that answers questions about me, streamed word by word.
  • Chat API validates input, rate limits per visitor and per site in Redis, and fails over from Gemini to Groq before the first word.
  • Next.js
  • TypeScript
  • Tailwind
  • Vercel AI SDK
  • Upstash Redis

view code ↗

// skills

Languages
  • Python
  • Go
  • C
  • C++
  • SQL
  • JavaScript
  • TypeScript
  • HTML/CSS
Frameworks
  • React
  • Next.js
  • Node.js
  • Flask
  • FastAPI
  • GraphQL
  • Ant Design
Tools
  • Git
  • Docker
  • Kubernetes
  • AWS
  • Terraform
  • Redis
  • GitHub Actions
Libraries
  • pandas
  • NumPy
  • Matplotlib
  • LangGraph
  • CrewAI
  • Jest
  • React Testing Library
Certifications
  • AWS Cloud Practitioner
  • AWS AI Practitioner
  • OCP Java SE 11

// contact

Open to internships and new-grad roles. The fastest way to reach me is email.

⎇ main
Toronto --:--