About Me
I'm a Machine Learning Engineer graduating from Michigan State University with a degree in Computational Data Science and a Mathematics minor. I specialize in building end-to-end AI systems — from data pipelines and model training to production deployment. My recent work includes a full RAG system for legal document research with citation verification, medical imaging classifiers published at SPIE, and Agentic-AI tooling for high-performance computing.
Core competencies:
- LLM applications, RAG pipelines, and retrieval systems
- Computer Vision and medical imaging (published researcher)
- Full-stack ML: Python, FastAPI, Docker, PostgreSQL, vector databases
- High-performance computing and distributed systems
- Production deployment and system architecture
I've contributed to published research at MIDI Lab (medical imaging) and built developer tools at MSU's Institute for Cyber-Enabled Research. I'm focused on building reliable, well-architected AI systems that deliver measurable results.
Education
B.S. Computational Data Science, Minor in Mathematics
Michigan State University
Spring 2026
Looking For
MLE / AI Engineer roles, ML consulting engagements, and research collaborations in NLP, computer vision, or applied AI.
My Skills
Programming Languages
ML & AI
Infrastructure & Web
My Resume
Professional Resume
View my comprehensive resume showcasing my education, experience, projects, and skills in Machine Learning and Data Science. Updated with my latest accomplishments and research work.
Resume PDF
View my complete professional resume
My Projects
End-to-end systems spanning ML pipelines, RAG architectures, and applied research
Legal AI — RAG Research Assistant
Open-source RAG system for legal and tax document research. Combines PostgreSQL full-text search with Qdrant vector similarity and a local LLM (Llama 3.1 via Ollama) to answer questions about uploaded documents. Every response passes through a citation verification layer that rejects claims not supported by retrieved evidence — making hallucinations structurally impossible. Supports multi-turn conversations, a full audit trail, and a structure-aware chunker that understands legal document formatting.
DarkVision
Computer Vision model for classifying animals in dark, low-visibility images with 92% accuracy using fine-tuned ResNet-18.
Auto Grader
Automated grading system that evaluates student code submissions against test suites, providing instant feedback and scoring.
Research & Publications
Ordinal Classification Framework for Multiclass Grading of Pneumoconiosis
Published research paper in SPIE Medical Imaging 2025 presenting a novel ordinal classification approach for automated pneumoconiosis severity grading.
This paper presents an ordinal classification framework specifically designed for multiclass grading of pneumoconiosis severity. Our approach addresses the inherent ordinal nature of pneumoconiosis progression stages, providing more accurate and clinically relevant automated assessment compared to traditional classification methods.
Published Paper
Access the publication online
DOI: 10.1117/12.3046353
Authors:
Liu, M., Loveless, I., Huang, Z., Borek, M., Rosenman, K., Alessio, A., Wang, L.
UURAF Research Poster 2025
Research poster presented at the University Undergraduate Research and Arts Forum (UURAF), showcasing AI-powered pneumoconiosis classification using chest radiographs.
Pneumoconiosis is an occupational lung disease caused by inhaling mineral dust, and chest radiography remains the key screening tool. Although standardization efforts by the ILO and NIOSH—such as the B Reader Certification Program—have improved consistency, challenges like reader variability, limited certified readers, and potential conflicts of interest persist. This study leverages artificial intelligence to objectively classify pneumoconiosis severity on a 4-point scale (0–3) using posterior-anterior chest radiographs from the NIOSH repository. A ResNet framework employing various loss functions (cross-entropy, corn, coral, focal staging, hierarchical, and hierarchical cross-entropy) is explored to enhance diagnostic reliability.
Research Poster
View or download the full poster
HPC Agentic-AI Framework for Batch Job Script Validation
Research poster presenting an innovative Agentic-AI framework designed to help HPC users validate batch submission scripts using large language models to reduce computational waste.
High-performance computing (HPC) users frequently make errors when writing batch job submission scripts, e.g. syntax errors, references to unavailable software installations and/or data files, inappropriate resource requests for the given cluster. These errors generally result in failed submissions or worse, jobs failing after having spent significant time in a queue or after having run for some time on the cluster, leading to wasted compute cycles with unnecessary energy consumption and needlessly prolonging the research cycle. This project aims to help HPC users increase efficiency and productivity by employing an HPC hosted large language model (LLM) as an Agentic-AI framework designed to examine batch submission scripts and advise users on potential errors in syntax, software/file refences, and resource allocation prior to submission. We first focus our efforts on the Michigan State University High-Performance Computing Center, using the 'codellama' family of LLMs.
Research Poster
View or download the full poster
Get In Touch
Let's Connect
I'm currently looking for new opportunities and collaborations in the field of Machine Learning. Feel free to reach out if you have any questions or just want to say hi!