page 2 · about
Hi, I'm Rajan.
I like the part of engineering where something leaves my laptop and other people start depending on it. Most of what I've built came from being annoyed at something at work: nobody knew what our agents cost, so I built Neurovn. Our agent configs didn't work across IDEs, so I built Rosetta. Claude Code kept eating our API budget, so Aditya and I built CacheLane.
At work I've bounced between the model side and the infrastructure side: shrinking and serving models on GPUs at RBC Borealis AI, writing Terraform at Intact, building a research agent at RBC that became a patent application, and now keeping deploys boring at Bree.
- based
- Toronto, ON · UTC−5 · open to US/Canada
- school
- B.Sc. Computer Science, Western University · May 2027 · 3.7 GPA
- now
- Software Engineer Intern at Bree (YC)
- off the clock
- markets (I used to day trade, badly, then less badly), hackathons, writing
Recognition
the fridge-magnet section- 2025Patent pendingagentic orchestration over regulated financial data, from my RBC Amplify internship
- 2026UofT Hacks, winnerBadge: vectorized professional identity for career fairs
- 2026Apple Swift Student ChallengePhysioPoint: ARKit physiotherapy assessment
- 2026Rosetta at Borealis AIpitched to the lab, picked up by engineers on the team
- ·Western AI, executiveled the CNN-based ASL translator team
- ·Western Design Thinking, executivepitched a GPS car-theft idea to the Dean of Engineering
on my desk
- The Bitter LessonRich Sutton
- Reflexion: Language Agents with Verbal Reinforcement LearningShinn et al.
- Designing Data-Intensive ApplicationsMartin Kleppmann
on the bench
- □distributed training internals (PyTorch FSDP, DeepSpeed)
- □graph algorithms for agent workflows (Tarjan's SCC and friends)
- □Triton kernels and CUDA fundamentals
checkboxes stay empty until I ship something with it
colophon
Next.js on Vercel. Set in Instrument Serif, Inter Tight, JetBrains Mono and Caveat for the handwriting. Graph paper by day, blueprint by night. Demo clips are real screen recordings. Find the source on GitHub.