Taanish R

hi,
I’m Taanish Reja, a 4th year CS student @ UCSD. I like working on compilers, PL and GPU programming problems.

you can find my github and linkedin here:

coding activities

work experience

Software Engineering Intern @ Ursa Major (Jun-Aug 2026)

Fixed a lot of React issues and worked on a cool interactive test replay viewer that colors different parts of a rocket engine CAD model over time based on the location of selected sensors and their readings.

CAD model viewer

a screenshot of my intern project, built on top of React and Three.js

Instructional Assistant for CSE 120 (Operating Systems) @ UCSD (Mar 2026-Present)

I’ve been an IA for my school’s operating systems course for two quarters. My responsibilities vary quarter to quarter, but I generally hold office hours, grade, and help the prof plan the course.

projects

Butterfly

Declarative, GPU-rendered UI library that implements a rough subset of HTML/CSS

Some cool demos made using my library:

Wikipedia

a wikipedia clone in my gui library

Other example

a demo claude generated on its own whims using my library

bookmarkmanga.com

Manga tracker w/ AI features; Next.js frontend + fastify backend + Rust inference + Some python services (e.g. faiss), hosted on AWS. It is currently down :(

Some screenshots:

home page

the home page

profile page

the profile page

notes page

the notes page

graph

the semantic graph built from users’ notes

stats page

the stats page

accelerarch.com

Project management software for architects; Next.js frontend + fastify backend, hosted on AWS

optimizing compiler for lisp-Like Language

Wrote an optimizing compiler for my compilers class in Rust. Ended up with dead-code elimination, constant folding, loop unrolling, some inlining, and JIT, AOT and REPL modes. The JIT took weeks off my life.

peer-to-peer vs code live share

For my distributed systems class, I built a peer-to-peer version of vs code live share. It’s based on a leader-replica model, where replicas send their edits to a leader that reorder them via a sequencing algorithm based on a lamport clock (per sequencing round) and wall-cllock time. I used raft for leader reelection, and employed one trick from Bayou where replicas preemptively sequence their own edits with respect to their last recieved log (not full-blown anti-entropy, though, as this project had a leader-replica model).

things unrelated to employment (this is a personal website):