Terrance Luangrath

I finished my B.S. in Mathematics with a minor in Data Analytics at James Madison University in 2024, then my M.S. in Data Science at the University of Virginia in August 2026. Most of what I build now runs on LLMs and agents: fine-tunes, tool-calling pipelines, retrieval, the pieces that decide whether a model is reliable enough to hand real work to. I still reach for classic ML where it's the better tool for the job, and I'm usually rewriting some part of a pipeline because I found a cleaner way to do it.

What I've done

• Shipped a course-recommendation agent covering 2,759 courses, 1,754 skills, and 3,682 prerequisite relationships, built with LangChain and a local Llama 3.1, packaged in Docker Compose.
• Led integration of a five-agent pipeline (ingest → normalize → score → publish → persist) that turns public packaging news into traceable market-by-technology opportunity scores, and built its LLM-based normalization agent (Claude Sonnet 5, with a self-hosted Qwen3-4B fallback) against a 55-segment market taxonomy and 16-category technology table. Capstone with Smurfit WestRock at UVA's School of Data Science.
• Validated 1,500+ records a day for a year as a Junior Data Analyst at ACTFORE, keeping data accuracy above 99%, June 2024 to May 2025.
• Built a fraud-detection pipeline that flags fraud across 6.3M+ transactions at 94% recall and 0.99 ROC-AUC — the hard part was the severe class imbalance. Deployed on Azure ML and Azure Container Apps.
• Co-authored a UVA Deep Learning benchmark of a CNN against an Audio Spectrogram Transformer for keyword spotting under noise; led the dataset, experiments, and spectrogram/spectral-gating implementation. CNN won clean audio 94.76% to 90.52%, and stayed ahead as noise got worse.

What I want next is AI engineering work: building the systems around LLMs, not just the models, and figuring out how you check that one got something right before anyone downstream trusts it. I'm job hunting for those roles, plus applied ML, right now. If any of this is useful to you, the form below reaches me.