Fullstack Software Engineer

RAJ DESAI

MS Computer Science · UT Dallas · 2026

DESIGN✦BUILD✦AUTOMATE✦SHIP✦SCALE✦REFINE✦DESIGN✦BUILD✦AUTOMATE✦SHIP✦SCALE✦REFINE✦DESIGN✦BUILD✦AUTOMATE✦SHIP✦SCALE✦REFINE✦DESIGN✦BUILD✦AUTOMATE✦SHIP✦SCALE✦REFINE✦

Exploring & building good sh!t since 2023.

Fullstack, product-minded, and comfortable owning a thing end to end. Four projects below, all shipped, all live.

What's below

Available Now

Full-time & contract, US-based.

GitHub LinkedIn

Scroll

I'm an engineer, builder, & designer always chasing the next thing worth making.

I grew up in Mumbai, moved to Dallas for my masters, and somewhere along the way got obsessed with building software that actually feels good to use. I think deeply about products, the thought behind them, and why they should exist in the first place.

Currently open to full-time roles starting July 2026.

Outside of engineering, building, and designing, I'm:

reading anything that makes me better at what I do
scrolling YouTube for what's new in tech and cars
discovering music nobody's heard of yet
always out with the camera, chasing skies and sunsets
finding new anime to binge

indie acoustics.Open in Spotify

1/4

FRONTEND

10

React

Next.js

TypeScript

Tailwind CSS

Framer Motion

GSAP

shadcn/ui

Recharts

Flutter

Figma

BACKEND

11

Node.js

FastAPI

PostgreSQL

Redis

BullMQ

Express.js

REST APIs

MongoDB

Prisma ORM

OAuth 2.0

JWT

AI / ML

9

LLMs

FAISS

OpenAI / Gemini APIs

Sentence Transformers

Semantic Search

Prompt Engineering

NLP

Scikit-learn

PyTorch

ARCHITECTURE

6

System Design

Async Pipelines

API Design

Real-time Data Flows

Multi-tenant Systems

Database Design

TOOLING

8

Git

Docker

GitHub Actions

CI/CD

Vitest

Claude Code

Cursor

Vercel

TOOLING

8

Git

Docker

GitHub Actions

CI/CD

Vitest

Claude Code

Cursor

Vercel

TOOLING

8

Git

Docker

GitHub Actions

CI/CD

Vitest

Claude Code

Cursor

Vercel

01

CLAUSEGUARD

Clause-by-clause contract risk analysis with redlines you can paste straight into a negotiation.

ClauseGuard takes an NDA, MSA, or SOW and returns a plain-English risk breakdown for every clause, plus a redlined Word document for the risky ones. Analysis runs as a five-stage Celery saga, and each stage commits before the next begins, so a worker can crash mid-run without losing paid work. Classification fans out one parallel task per clause. Every model call passes a circuit breaker with automatic provider failover, and every clause hits a cache before it is allowed to cost anything. Nine containerized services, deployed and running on a VPS I administer myself.

Next.jsTypeScriptFastAPICeleryRabbitMQPostgreSQLpgvectorRedisDockerLiteLLM

Live Demo GitHub

I wanted one project where the hard part was not the feature list. Anything can call an API and render the response. Keeping it correct is the actual work: what happens when a provider degrades, when a worker dies at stage four, when the same clause arrives for the two hundredth time, when someone needs their data back. I also wanted to find out whether I could run the whole thing myself rather than handing the interesting parts to a platform.

02

YORU

A full-screen ambient station where you don't press play, you step inside.

Yoru is a full-screen listening environment built around 21 anime-style scenes and 13 layered soundscapes. Scenes and sounds are mapped together -- switching audio automatically transitions the visuals to match. The UI fades away after a few seconds of inactivity so there's nothing between you and the atmosphere. Every transition crossfades, every interaction has keyboard support, and the whole experience is designed to make you forget you're in a browser.

Next.jsReactTypeScriptTailwind CSSFramer MotionHowler.jsshadcn/uiVitest

Live Demo GitHub

I wanted to build something where the goal wasn't productivity or metrics -- just a feeling. Most music sites are built around playlists and controls. I wanted to see what happens when you strip all of that away and design around atmosphere instead.

03

PR SENSEI

AI-powered code reviews that land on your PR in under 30 seconds.

···

PR Sensei hooks into GitHub's webhook system to intercept pull requests, queues them through an async Redis and BullMQ pipeline, and generates structured code reviews using OpenAI and Gemini APIs -- all in under 30 seconds. It posts a summary comment and up to 5 inline comments on the exact changed lines. Reviews are deduplicated so the same line never gets flagged twice, and the whole system is idempotent per commit SHA. A Next.js dashboard tracks review metrics, file hotspots, and history across repos, backed by a multi-tenant PostgreSQL schema.

Next.jsTypeScriptNode.jsFastifyRedisBullMQPostgreSQLPrisma ORMOpenAI APIGemini APIOAuth 2.0

GitHub

Code reviews are a bottleneck on every team I've been on. Someone opens a PR, and it sits there for hours waiting for a human to look at it. I wanted to build something that gives developers fast, structured feedback the moment they push -- not to replace human review, but to catch the obvious stuff so the real conversation can focus on architecture and design.

04

LLM COOKBOOK

Type in what's in your fridge, get back recipes that actually make sense.

···

LLM Cookbook matches natural-language queries to recipes using FAISS vector search and MiniLM embeddings -- so searching "something warm and spicy" returns results by meaning, not keywords. If nothing matches well enough, it falls back to a local LLM to generate a recipe from scratch. The system handles dietary filters, allergen exclusions, and ingredient substitution. A Pandas pipeline processes 10,000+ raw recipe records into model-ready embeddings in a single reproducible script, served through a FastAPI backend.

PythonFastAPIFAISSSentence TransformersOllamaOpenAI APIStreamlitPandasNumPy

GitHub

I wanted to see how far I could push semantic search before needing a full LLM. Recipes were the perfect domain -- structured enough to test retrieval quality, messy enough to need real NLP. It started as a class project and turned into a deep dive into embeddings, vector search, and knowing when to let the model generate vs when retrieval is enough.

ClauseGuard takes an NDA, MSA, or SOW and returns a plain-English risk breakdown for every clause, plus a redlined Word document for the risky ones. Analysis runs as a five-stage Celery saga, and each stage commits before the next begins, so a worker can crash mid-run without losing paid work. Classification fans out one parallel task per clause. Every model call passes a circuit breaker with automatic provider failover, and every clause hits a cache before it is allowed to cost anything. Nine containerized services, deployed and running on a VPS I administer myself.

Next.jsTypeScriptFastAPICeleryRabbitMQPostgreSQLpgvectorRedisDockerLiteLLM

Live Demo GitHub

I wanted one project where the hard part was not the feature list. Anything can call an API and render the response. Keeping it correct is the actual work: what happens when a provider degrades, when a worker dies at stage four, when the same clause arrives for the two hundredth time, when someone needs their data back. I also wanted to find out whether I could run the whole thing myself rather than handing the interesting parts to a platform.

Yoru is a full-screen listening environment built around 21 anime-style scenes and 13 layered soundscapes. Scenes and sounds are mapped together -- switching audio automatically transitions the visuals to match. The UI fades away after a few seconds of inactivity so there's nothing between you and the atmosphere. Every transition crossfades, every interaction has keyboard support, and the whole experience is designed to make you forget you're in a browser.

Next.jsReactTypeScriptTailwind CSSFramer MotionHowler.jsshadcn/uiVitest

Live Demo GitHub

I wanted to build something where the goal wasn't productivity or metrics -- just a feeling. Most music sites are built around playlists and controls. I wanted to see what happens when you strip all of that away and design around atmosphere instead.

···

PR Sensei hooks into GitHub's webhook system to intercept pull requests, queues them through an async Redis and BullMQ pipeline, and generates structured code reviews using OpenAI and Gemini APIs -- all in under 30 seconds. It posts a summary comment and up to 5 inline comments on the exact changed lines. Reviews are deduplicated so the same line never gets flagged twice, and the whole system is idempotent per commit SHA. A Next.js dashboard tracks review metrics, file hotspots, and history across repos, backed by a multi-tenant PostgreSQL schema.

Next.jsTypeScriptNode.jsFastifyRedisBullMQPostgreSQLPrisma ORMOpenAI APIGemini APIOAuth 2.0

GitHub

Code reviews are a bottleneck on every team I've been on. Someone opens a PR, and it sits there for hours waiting for a human to look at it. I wanted to build something that gives developers fast, structured feedback the moment they push -- not to replace human review, but to catch the obvious stuff so the real conversation can focus on architecture and design.

···

LLM Cookbook matches natural-language queries to recipes using FAISS vector search and MiniLM embeddings -- so searching "something warm and spicy" returns results by meaning, not keywords. If nothing matches well enough, it falls back to a local LLM to generate a recipe from scratch. The system handles dietary filters, allergen exclusions, and ingredient substitution. A Pandas pipeline processes 10,000+ raw recipe records into model-ready embeddings in a single reproducible script, served through a FastAPI backend.

PythonFastAPIFAISSSentence TransformersOllamaOpenAI APIStreamlitPandasNumPy

GitHub

I wanted to see how far I could push semantic search before needing a full LLM. Recipes were the perfect domain -- structured enough to test retrieval quality, messy enough to need real NLP. It started as a class project and turned into a deep dive into embeddings, vector search, and knowing when to let the model generate vs when retrieval is enough.

···

Jul 2026 – Present

Full-Stack Software Engineer

Universal Removal Defense Project

Volunteer engineer on an open-source Django platform routing immigrants and asylum seekers to licensed immigration attorneys.

→Co-own the intake platform as one of two engineers on the repo, serving 10 languages including right-to-left scripts

→Audited architecture, request flow, and code quality before contributing, then converted the open backlog into a sequenced implementation plan the team is executing against

→Restored multilingual support by tracing a silent locale-compilation failure across Docker startup paths

Sep 2025 – May 2026

Student Assistant

University of Texas at Dallas

Accessibility support and front-desk operations at UT Dallas's AccessAbility Resource Center.

→Scribe and read notes for visually challenged students, translating code, mathematical equations, and technical concepts into accessible formats

→Built a Python script with AI to convert handwritten mathematical notes into JAWS-compatible formats (.docx, .html)

→Manage exam proctoring to maintain a fair and accessible testing environment for students with disabilities

→Handle front-desk operations, student and faculty inquiries, and administrative support

Jan 2023 – Dec 2023

Software Engineer Intern

Hydracoral Technologies

Full-stack development across Flutter mobile and Django web for a pre-launch product in a 3-person Agile team.

→Developed a Flutter/Dart onboarding flow with facial-recognition APIs, cutting user verification time by 30% in staging

→Diagnosed and resolved 15+ bugs in state management, API error handling, and memory leaks using Firebase Crashlytics -- reducing crash rate by 20% pre-launch

→Designed responsive cross-platform UI components validated through structured user testing, improving usability scores by 20%

→Owned CI/CD pipelines, integrating REST APIs for JWT authentication, data sync, and release automation

Jul 2026 – Present

Full-Stack Software Engineer

Universal Removal Defense Project

Volunteer engineer on an open-source Django platform routing immigrants and asylum seekers to licensed immigration attorneys.

→Co-own the intake platform as one of two engineers on the repo, serving 10 languages including right-to-left scripts

→Audited architecture, request flow, and code quality before contributing, then converted the open backlog into a sequenced implementation plan the team is executing against

→Restored multilingual support by tracing a silent locale-compilation failure across Docker startup paths

Sep 2025 – May 2026

Student Assistant

University of Texas at Dallas

Accessibility support and front-desk operations at UT Dallas's AccessAbility Resource Center.

→Scribe and read notes for visually challenged students, translating code, mathematical equations, and technical concepts into accessible formats

→Built a Python script with AI to convert handwritten mathematical notes into JAWS-compatible formats (.docx, .html)

→Manage exam proctoring to maintain a fair and accessible testing environment for students with disabilities

→Handle front-desk operations, student and faculty inquiries, and administrative support

Jan 2023 – Dec 2023

Software Engineer Intern

Hydracoral Technologies

Full-stack development across Flutter mobile and Django web for a pre-launch product in a 3-person Agile team.

→Developed a Flutter/Dart onboarding flow with facial-recognition APIs, cutting user verification time by 30% in staging

→Diagnosed and resolved 15+ bugs in state management, API error handling, and memory leaks using Firebase Crashlytics -- reducing crash rate by 20% pre-launch

→Designed responsive cross-platform UI components validated through structured user testing, improving usability scores by 20%

→Owned CI/CD pipelines, integrating REST APIs for JWT authentication, data sync, and release automation

2024 – 2026

M.S. Computer Science

University of Texas at Dallas

Focused on going deeper into AI/ML systems and strengthening core fundamentals.

→Coursework in Machine Learning, NLP, AI, Big Data Analytics, Database Design, and Design & Analysis of Algorithms

→Focused on integrating AI/ML into full-stack products -- from semantic search pipelines to LLM-powered developer tools

→Strengthened core CS fundamentals in algorithms, systems design, and data modeling

2020 – 2024

B.Eng. Computer Engineering

University of Mumbai

Built a foundation in computer science fundamentals -- data structures, algorithms, operating systems, databases, and software development.

→Chairperson, CSI-TCET -- managed 80 committee members, organized technical seminars, hackathons, a 3-day technical festival, and a 150+ student out-of-state industrial visit

→Deputy Cultural Secretary, TSDW TCET -- organized the college's 3-day cultural festival SOJOURN

→Art Designer Lead, Nimbus -- led a 4-person team designing the department magazine from concept to launch

→Developed early interest in fullstack engineering and product thinking that carried into everything built after

2024 – 2026

M.S. Computer Science

University of Texas at Dallas

Focused on going deeper into AI/ML systems and strengthening core fundamentals.

→Coursework in Machine Learning, NLP, AI, Big Data Analytics, Database Design, and Design & Analysis of Algorithms

→Focused on integrating AI/ML into full-stack products -- from semantic search pipelines to LLM-powered developer tools

→Strengthened core CS fundamentals in algorithms, systems design, and data modeling

2020 – 2024

B.Eng. Computer Engineering

University of Mumbai

Built a foundation in computer science fundamentals -- data structures, algorithms, operating systems, databases, and software development.

→Chairperson, CSI-TCET -- managed 80 committee members, organized technical seminars, hackathons, a 3-day technical festival, and a 150+ student out-of-state industrial visit

→Deputy Cultural Secretary, TSDW TCET -- organized the college's 3-day cultural festival SOJOURN

→Art Designer Lead, Nimbus -- led a 4-person team designing the department magazine from concept to launch

→Developed early interest in fullstack engineering and product thinking that carried into everything built after

Got something interesting to build, discuss, or explore?

I'd love to hear from you.