Work Experience

Supporting bar and service operations at a family-owned Irish pub in downtown Troy, keeping bars stocked, glassware clean, and service running smoothly.

• Restocked bar supplies and drinks into coolers throughout each shift to keep service running without interruption.
• Washed and sanitized pint glasses to maintain a steady supply of clean glassware behind the bar.
• Delivered food orders to tables, supporting servers during busy periods.
• Retrieved and restocked fresh ice for the bars on a continuous basis.
• Cleaned and reset the venue after private events to prepare for regular service.
• Refilled drink stations as needed to keep bartenders stocked during peak hours.

Building AI-powered search and evaluation tools for a multilingual mental-health knowledge platform, helping researchers find reliable information more efficiently.

• Designed a content-addressed caching pipeline (SHA-256 manifests) that deduplicates and incrementally re-embeds a 550+ item, 28-language knowledge corpus, eliminating redundant embedding-API calls.
• Architected a pluggable vector store abstraction spanning four similarity metrics that fails over transparently between a Milvus cluster and an in-process NumPy backend, achieving zero query failures during outages.
• Built an MCP tool server and FastAPI REST service on a shared retrieval backend, extending semantic search to both autonomous AI agents and conventional applications through one swappable interface.
• Developed an automated evaluation harness benchmarking ranking accuracy across embedding models and similarity metrics, producing versioned reports for every corpus revision.

Related public work: POEM ontology project at Tetherless World. The evaluation-harness bullet above is demonstrated directly in PR #13 (expanded the eval suite from 12 to 50 queries across every ontology dimension), and the retrieval-pipeline bullets build on PR #11.

Worked closely with a professor and a small research team to build and test AI language models, troubleshooting problems together and coordinating across the team to bring reliable systems into real-world use.

• Boosted scientific reasoning benchmark performance 15% by fine-tuning Gemma 7B and Qwen 7B with parameter-efficient training and hyperparameter optimization.
• Cut production inference latency below 800ms in deployed voice-AI systems by optimizing streaming inference and tool-use orchestration.
• Built evaluation infrastructure comparing accuracy, robustness, latency, and failure modes across competing LLM architectures to guide deployment decisions.

Related public work: a real fine-tuning-pipeline commit to FoamGPT, Prof. Shaowu Pan's OpenFOAM/LLM research project at RPI. This project led to a published workshop paper by the lab: “FoamGPT: Fine-Tuning Large Language Model for Agentic Automation of CFD Simulations with OpenFOAM” (NeurIPS 2025 ML4PS workshop) and its arXiv companion, “Foam-Agent”, noted here as resulting work from the lab, not as an authorship credit.

Taught and supported undergraduates in Multivariable Calculus and Matrix Algebra.

• Held weekly office hours and pre-exam review sessions and graded assignments for 2+ undergraduate courses, delivering regular feedback to support student understanding.

See also: B.S. Computer Science & Mathematics, RPI.

Developed mesh-adaptation software in C++ to improve CFD simulation accuracy for unsteady-state systems.

• Integrated mesh-adapt tools with the CREATE-FT Capstone SDK and Raven CFD, applying differential equations and numerical methods to automate mesh refinement near critical flow features for capsule-reentry simulations.

Research for RPI's DMREF (“Designing Materials to Revolutionize and Engineer our Future”) project, applying deep learning — autoencoders and graph neural networks — to material properties for aircraft-engine components under extreme conditions, using the Materials Project and AFLOW databases with RPI's Mechanical, Aerospace, and Nuclear Engineering department.

• Engineered deep autoencoder architectures and ran systematic experiments across model configurations, improving telemetry prediction performance on holdout datasets.
• Implemented and benchmarked four graph neural network models — a GCN, a GAT, and Graph Autoencoders (GAE) built on each — trained with 5-fold cross-validation; the GCN alone hit R² = 0.971 on the Materials Project dataset.
• Established reproducible benchmarking pipelines and evaluation frameworks that increased the stability and validity of experiment results.
• Automated training, validation, and performance-analysis workflows, accelerating iteration speed and surfacing latent-space representations that improved downstream predictive performance.
• Earlier phases swept autoencoder bottleneck-layer sizes on a 15-feature aircraft-material dataset and explored a separate Keras/TensorFlow CNN autoencoder over image-based material representations; later work applied the pipeline to a custom dataset in collaboration with Dr. Subrato's team.

Related public work: Graph Neural Networks for Material Property Prediction (repo & result images on the Projects page), applying graph neural networks and autoencoders to jet-engine materials. This work also received RPI's 2024 Summer Undergraduate Research Award.

Taught AI, machine learning, and Python programming courses to students ages 10–17.

• Designed lesson plans and assessments for weeklong AI/ML and Python courses while overseeing the safety and experience of residential campers.

⤢

← Prev 1 / 2 Next →

Researched and prototyped an engineering solution as part of a competitive STEM internship cohort.

• Collaborated with a team of 4, logging 20+ hours a week to research and present an engineering solution to Boeing engineers, earning the program's “Best Idea” award among 17 teams.

Applied and was selected to serve as a Virtual Learning Ambassador for the Hawthorne Unified School District's YMCA after-school program.

• Completed a 48-hour training program covering Job Readiness, Child Abuse Awareness, and G-Suite Certification.
• Assisted students who needed technical assistance while in distance-learning mode.
• Researched and created job listings for high schoolers.