Project Portfolio
Below is a curated selection of applied projects spanning AI systems, scientific production environments, and infrastructure design. These efforts reflect a consistent focus on ownership, system-level thinking, and building things that operate reliably in the real world.
🏠 HomeRepairBot — Multimodal Vision-RAG Assistant
HomeRepairBot is a production-oriented AI system designed to assist users with diagnosing and resolving common home repair and maintenance tasks. The system integrates computer vision, natural language processing, and retrieval-augmented generation to produce structured, actionable guidance rather than generic responses.
I designed and built the backend inference pipeline, safety and intent gating logic, session memory handling, and document retrieval flow. The system supports local inference, curated knowledge sources, and multi-turn interactions, with a mobile frontend used to capture images and user context.
⚗️ Iodine Electrochemical Cell — Production System Development
I led the design, scale-up, and deployment of an iodine-based electrochemical production system, translating bench-scale research into a functioning production environment. This work required integrating electrochemistry, instrumentation, automation, and safety considerations into a cohesive system.
Responsibilities included process design, sensor integration, control logic, troubleshooting, and authoring operating and maintenance documentation. I coordinated between R&D, operations, and management to ensure the system met performance, safety, and regulatory requirements.
🖥️ HomeLab Network — Compute, Storage, and Services
I designed and maintain a self-hosted HomeLab network that serves as both a development environment and a deployment platform for AI workloads, data pipelines, and networked services. The system includes a multi-boot workstation and a Debian-based server hosting containerized applications.
Services include private email, web hosting, VPN access, media services, and AI inference backends, all secured with custom networking and firewall configurations. This environment functions as my primary lab for experimentation, testing, and long-running services.
🎥 MOT-ReID — Multi-Object Tracking & Re-Identification
This project focused on building a pedestrian tracking pipeline that combines object detection with deep re-identification to maintain identity consistency across frames and camera views. The system integrates multiple deep models and evaluates performance using standard tracking metrics.
As project lead, I managed model selection, training workflows, evaluation, and repository structure, and presented results and trade-offs during reviews. The work emphasized reproducibility, quantitative evaluation, and clear communication of system behavior.
🤖 AI / ML Backbone — Applied Methods & Tooling
Across a range of smaller projects and experiments, I have deployed both classical and modern machine-learning techniques to real datasets and systems. Rather than focusing on single showcases, this work reflects a bundled skillset applied repeatedly across domains.
In computer vision, I have implemented classic feature-based approaches alongside deep learning models, including custom embedding pipelines for downstream tasks. In NLP, I have built preprocessing, tokenization, and classification pipelines using both traditional and neural approaches. I have also trained and fine-tuned local models, emphasizing evaluation, trade-offs, and resource-aware deployment.
🔥 Biomass Gasifier — Performance Optimization
I contributed to improving the efficiency of a prototype biomass gasifier through experimental diagnostics and data-driven analysis. Working alongside engineers and technicians, I helped evaluate combustion behavior and interpret laboratory testing results.
The findings informed adjustments to system configuration and operating parameters, resulting in a measurable improvement in throughput and performance. This work reinforced the importance of combining theory, experimentation, and iteration in applied energy systems.
🚀 Current & Future Work
Ongoing efforts center on expanding the HomeLab Network with additional compute nodes, including a multi-AMD powered AI rack intended for model deployment, fine-tuning, and experimental workloads.
Future directions also include deeper evaluation tooling for AI systems, expanded deployment of multimodal services, and continued exploration of system architectures that support reliable, interpretable intelligence in real environments.