A scalable, distributed web crawler and RAG pipeline built on my own extendable worker framework where nodes communicate via Redis queues. It features secure, localized text processing—including chunking and embedding inference—and autoscaling of scraper and RAG clusters based on custom Redis queue metrics refreshed every minute by AWS Lambda. The system also reliably extracts email addresses, phone numbers, and links from web pages while maintaining very high test coverage.

Open sourced from the MVP of my startup, Mia Marketing. Features an entire pipeline of generating tailored marketing campaigns for any business on any topic, generated from just their URL. Uses extensive web scraping and LLM manipulation. Integrated with Canva for customizing AI-generated templates. Also features a dashboard for interacting with the application.

After noticing a pump and dump scheme involving botted wallets on the Pumpfun platform, I built a bot that would track down these bots, and could copy trade them and then sell before the bots would sell, making profit.

Mobile app for university students to find project groups, inspired by Tinder, featuring a custom matching algorithm, project creation, recommendations, and integrated chat rooms. Collaborated with stakeholders to define and refine requirements, ensuring user needs alignment.

Full stack dashboard enabling customer profiling, bookings and venue management. Includes a content management system for populating websites. Open-sourced from my startup.

AArch64 emulator that simulates the execution of an AArch64 binary file on a Raspberry Pi, and AArch64 assembler that translates an AArch64 assembly source file containing A64 instructions into a binary file that can subsequently be executed by the emulator.

Full stack Trello clone with authentication, organizations, workspaces, stripe subscriptions and CRUD operations implemented with server actions.

Airbnb mobile app clone, developed with React Native.

Built a blisteringly fast AI agent framework from scratch in Go after being unsatisfied with the existing frameworks. Features easy switching of LLMs, memory, tool calling, and RAG. Used this framework to build a chain of thought ReAct agent capable of replacing a personal tutor for a university student, at the request of a client. Discovered an intuitive way to implement the Think, Act, Observe loop (with chain of thought) as a single LLM call.

A system which is able to watch any stream (TikTok, Twitch, Kick, TikTok, Youtube etc.) in real time, detect 'clippable moments', and then edit them into short form TikTok style videos. The plan was to sell a subscription to these clips as a service to clippers, but I had to abandon this as it was esentially serving copywrited content (so illegal)

Compiler for the WACC programming language that translates WACC source code into x86-64 machine code, handling lexical analysis, semantic analysis, code generation, and optimization. Web-based IDE built with Monaco editor and Next.js, supporting WACC code writing, compilation, and execution with syntax and error highlighting.

Large language model from scratch, trained on the OpenWebText dataset. Capable of generating human-like text using next token prediction.

MVP for my idea of a proximity-based social media mobile app. Profiles of nearby app users dynamically pop up on users' phones as they walk past each other, facilitating real-time interactions.

RSS feed web scraper with a REST API and authentication using api keys and middleware. Stores scraped results in a postreSQL database.

Deep multilayer perceptron neural network for handwritten digit recognition, without any machine learning or maths libraries. Implemented backpropagation from scratch.