AI Engineer
I build AI systems that do real work: autonomous coding agents, RAG search engines, and deep-learning pipelines taken from raw data to deployed model. Right now I'm focused on LLM agents and retrieval systems that actually ship — this site is one of them, rebuilding itself nightly from my GitHub.
I'm an AI engineer focused on the practical end of the field — getting agents, retrieval systems, and neural networks to actually ship. I've built LLM agents that write and execute code, retrieval-augmented search over real document corpora, an automated support-ticketing system, and reinforcement-learning agents that play Super Mario. On the deep-learning side I've trained and compared U-Net, ResNet, and VGG architectures for medical image segmentation, and worked on behavioral-biometric identification. This site is itself a small example of how I like to work: it rebuilds itself nightly from my GitHub account via a scheduled pipeline, ranks my repos automatically, and summarizes them with an LLM — zero manual upkeep.
The Automated Ticketing System is an AI-driven platform that manages IT support tickets by classifying issues and either resolving them automatically or escalating them for human review through a conversational interface. Technically, it features a robust backend built with FastAPI, integrates LLM for intelligent ticket processing, and employs a multi-turn chat system to maintain context during ticket creation and review.
This project is a machine learning system that identifies users based on their motion behavior captured by mobile device sensors, achieving a cross-validated accuracy of 94.7% with a Random Forest model. Notably, it employs extensive feature engineering, extracting around 609 features from sensor data and selecting the top 200 for model training, which enhances its ability to distinguish between users effectively.
The Interview Coach project provides personalized AI-driven interview practice by allowing users to upload their CV and project documents, input a job description, and engage in a tailored interview simulation. It is technically notable for its use of a local large language model (LLM) running on a GPU, combined with a FastAPI backend and a React frontend, ensuring that all data remains on the user's machine.
This project implements brain tumor segmentation using deep learning models, specifically U-Net, ResNet, and VGG architectures, on the BRaTS 2020 dataset. It is technically notable for its integration of a web application built with Flask, allowing medical practitioners to easily utilize the segmentation results generated by the models.
This project is a self-updating portfolio website that automatically regenerates its content from the GitHub API using GitHub Actions, ensuring it is always current without manual edits. Notably, it employs a pure function for ranking projects based on various criteria and integrates a fallback mechanism for generating project summaries, which enhances reliability and user experience.
The Coding Agent is a local tool that interacts with a workspace by inspecting files, making edits, and executing Python scripts, all while leveraging a llama.cpp model server for OpenAI compatibility. Notably, it features a modular architecture that integrates a custom MCP server for workspace management and optional tracing capabilities for enhanced workflow monitoring.