home@jellyware:~$ 

I previously worked at:

  • Meta - Software Engineering Intern (Facebook Messenger Web) (2025)
    • I worked on adding new features to Messenger & Facebook Messaging web to increase message sends (I can't share the features I worked on yet 😢)
    • Involved in both frontend and backend development of features - I worked with React & JavaScript
  • OncoFlow - Full-stack engineer (2024)
    • Joined as the first Full-Stack Engineer, leading the development of the company's main product.
    • Built a dashboard platform enabling NHS workers to log in, interface with and run AI models, connect to medical databases with cancer treatment information, and generate and edit AI-produced medical reports.
    • I had to design the frontend, architect the backend and build the app to a secure and reliable standard - taking the app from idea to testing and deploying it on a VPS.
    • I worked with Next.js, Typescript and Python

Read more about what I've done in my CV

Skills

Proficient

Python logoPython
|
TypeScript logoTypeScript
|
JavaScript logoJavaScript
|
C++ logoC++
|
React logoReact
|
Git logoGit
|
GitHub logoGitHub

Intermediate

C logoC
|
SQL logoSQL
|
Rust logoRust
|
Verilog logoVerilog

Cloud tech I have deployed apps and backends on

AWS logoAWS
|
Supabase logoSupabase
|
Vercel (Hosting this 😀) logoVercel (Hosting this 😀)

I'm also familiar with

Bash & Unix terminal logoBash & Unix terminal
|
Docker logoDocker
|
Figma (Proficient) logoFigma (Proficient)

Projects

VisuMath

VisuMath

  • This project won the research category of IC HACK 2025, Europe's largest student run hackathon!
  • This tool takes a math topic as input and generates a video (based on 3blue1brown videos) and interactive page.
  • Videos are stored on AWS S3 buckets, metadata is stored in dynamodb and video title are stored in a vector database for querying.
  • The backend consists of three layers: the API server (built with FastAPI, interfacing the video and meta data storage), a Redis message queue, and a Celery worker (runs the AI Agent workflow).
  • This architecture allows the API server to remain responsive while long-running tasks are offloaded to a separate process - generating a video takes around 2-5mins.
View CodeWatch the demo
snipr

snipr

  • Production-ready desktop app that removes silences and provides transcripts for audio files.
  • Currently has 1 user 😱
  • Implemented decoding and encoding audio files to and from PCM samples with FFMPEG, an algorithm to remove silences in O(n) time and provided a responsive UI by spawning new threads to process the audio file, all in Rust.
  • Implemented a CI-CD pipeline using GitHub Actions to build the app and provide automatic updates for users.
  • Developed a Rust-Python integration using PyO3 to call OpenAI’s Whisper model to provide audio transcriptions.
View Code
Rx2Label

Rx2Label

  • Checks a picture of a doctors prescription against medical literature (it can read a doctors handwriting :))
  • Takes a valid prescription and generates a label for the medication
  • Built at Hack UK hosted by a16z with Mistals Pixtral model
  • I Built the entire frontend and backend using React.js, Typescript, Tailwind and Python FastAPI
  • Supabase is used for image storage
View CodeWatch the demo
This site

This site

  • Built using Next.js, React.js, Typescript & SCSS
  • Deployed to Vercel and domain is protected by Cloudflare
  • CV is an API endpoint that serves my CV
  • Terminal animation at the top is built from scratch with React and SCSS
View Code
Jellis

Jellis

  • Implementation of C++ in Redis
  • Supports multiple clients simultaneously - implemented using the boost ASIO library with a event loop and TCP sockets
  • Developed a parser and encoder for the Redis serialization protocol (RESP)
  • Currently only supports the in memory database but I’m working on data persistence with RDB.
View Code
Neural Network from scratch - using Python and NumPy

Neural Network from scratch - using Python and NumPy

  • Implemented backpropagation to find the gradient of the loss with respect to each neuron
  • Implemented gradient descent to update the weight and reduce the loss
View Code
C90 to RISC V assembly - C++

C90 to RISC V assembly - C++

  • Achieved 89% - passing 179 out of 201 tests
  • Added support for: chars, ints, floats (IEEE 754), arrays, pointers, for, while & do while loops, if statements, switch statements, functions, strings, typedef keyword, arithmetic (pre-increment operator, post-increment, etc.), constants, local & global variables (including correct scoping), enums, sizeof
View Code
RISC V 32I Pipelined CPU - System Verilog & C++

RISC V 32I Pipelined CPU - System Verilog & C++

  • Designed and implemented the hardware for pipelining the CPU
  • Designed and implemented a hazard unit that could stall, flush instructions, and pass data to previous stages to prevent stalls
  • Wrote test benches in C++ to run programs on the CPU
View Code
mini grid project

mini grid project

  • Create a full-stack app with Next.js, Typescript, Tailwind CSS and Python
  • Displayed real-time data from Raspberry Pi Picos using MQTT
  • Programmed the Raspberry Pi Picos to communicate via MQTT
  • Used MongoDB to store previous data and plot it using Recharts.js
  • Deployed the MQTT broker to an AWS EC2 instance and shared the web app using ngrok
  • Built a neural network using PyTorch to predict the buy price of electricity