The concise version of the work.

Ames, IA / Expected Graduation: May 2028

• Bachelor of Science in Software Engineering
• Minors in Artificial Intelligence and Data Science

Jun 2026 - Present / Ames, IA

• Building computer vision pipelines for agricultural phenotyping, converting drone- and phone-recorded RGB tassel videos from field data collection into 3D point clouds for maize analysis.
• Supporting Structure-from-Motion and NeRF-based reconstruction workflows using COLMAP, Nerfstudio, OpenCV, Open3D, CloudCompare, CUDA, and Jetstream2.
• Developing data-processing pipelines for raw video, frame extraction, camera pose estimation, NeRF training, point-cloud export, and individual tassel .ply extraction.

Aug 2025 - Present / Ames, IA

• Expanded a D-ICL benchmark for tabular foundation models by adding 5 public regression datasets, large synthetic regression tasks, and OpenML Yolanda dataset 42705 to the evaluation pipeline.
• Ran TabPFN and TabICL experiments across IID/non-IID partitions, 120k-sample large-regression settings, and paper-aligned seeds, reporting RMSE, MAE, R2, and mean/standard-deviation summaries.
• Implemented batched regression inference to resolve CUDA memory limits on large test sets and generated reproducible JSON/CSV summaries supporting a paper currently under review at NeurIPS.

Jan 2025 - Jul 2025 / Ames, IA

• Developed a Python computer-vision pipeline using NumPy, Pandas, SciPy, OpenCV, and scikit-learn to detect and track spatter ejections in LPBF, improving tracking accuracy by approximately 20% over manual methods.
• Applied feature extraction and statistical modeling to analyze spatter velocity, size, and ejection angle across LPBF high-speed imaging experiments.
• Visualized 30,000 fps high-speed imaging data with Matplotlib, supporting reproducible large-scale ML experiments.

Rust / React / TypeScript / WebSockets / SQLite / Jun 2026

• Built a full-stack LPBF monitoring app with real-time telemetry, recipe sequencing, rule-based anomaly detection, SQLite-backed run history, and process simulations for oxygen, temperature, recoater, laser, and spatter behavior.
• Developed an async Rust/Axum/Tokio backend with REST APIs, SQLx, and 300 ms WebSocket updates, plus a React/TypeScript dashboard with live charts, alerts, run controls, experiment history, and automated tests.
• Repository: github.com/somtri/run_scope.git

Python / Random Forest / Streamlit / Dec 2025

• Built a Random Forest stock-direction pipeline using 26 engineered price, volume, volatility, momentum, and sentiment features with automated yfinance ingestion, preprocessing, and model persistence.
• Implemented leakage-aware chronological holdout and expanding-window validation, achieving 63.3% walk-forward accuracy on a deterministic market simulation, with Streamlit dashboards for ROC AUC, equity curves, confidence, and feature importance.
• Repository: github.com/somtri/smart_signal.git

Software / ML / Research

• Python, Rust, TypeScript/JavaScript, SQL, Java, C/C++

Research systems / application development

• PyTorch, scikit-learn, NumPy, Pandas, OpenCV, Open3D, COLMAP, Nerfstudio, CloudCompare, React, Streamlit, Plotly, Axum, Tokio, SQLx, SQLite, yfinance, Git, Linux, CUDA, Jetstream2, pytest, GitHub Actions

Methods

• Machine Learning, Computer Vision, Tabular Foundation Models, Structure from Motion, NeRF, Time Series Forecasting, Feature Engineering, WebSockets, REST APIs, Batched Inference, Anomaly Detection, Reproducible ML Experiments