Research rooms.
Production systems.

Advisor: Ray Friedman

• I led end-to-end LLM research for negotiation science, from model design and experimentation through evaluation and deployment.
• I built an LLM pipeline that automatically codes negotiation transcripts, reducing the per-transcript cost from $5,000 to $3, a savings of more than 99%, while increasing human–AI agreement from 30% to 80%.
• I developed an algorithm that introduces population-level variation into LLM negotiator agents, producing more realistic and diverse subject pools for negotiation studies.
• I worked closely with negotiation scholars and data science researchers across data engineering, modeling, and web deployment.
• The work was presented at the 2025 AI Negotiation Summit at Harvard and MIT and at the International Association for Chinese Management Research annual meeting.

Advisor: Tyler Derr

My research focused on a unified multimodal fusion framework spanning vision, graph, and language modalities.

• Vision–Language: Designed Inverse-LLaVA, a single-stage architecture that maps intermediate text representations into continuous visual space. It uses zero alignment pre-training samples and 45% fewer total training samples than LLaVA-1.5 while remaining competitive across nine benchmarks, with a 27.2% higher cognition score on MME.
• Graph–Text: Implemented a Temporal Graph Attention Network (TGAT)–LLM pipeline that generates high-quality textual attributes for social networks. Evaluated the approach on the Venmo Dataset and designed a follow-up evaluation for the Amazon Review Dataset.

I supported three courses through a Graduate Teaching Fellowship, organizing labs, creating quizzes, grading coursework, and hosting office hours.

Graduate courses:

• DS 5620: Probability and Statistical Inference
• DS 5690: Gen AI in Theory and Practice

Undergraduate course:

• DS 3100: Fundamentals of Data Science

Advisor: Markus Eberl

• I improved ancient mortar classification accuracy from 60% to 97% by training Vision Transformers on 10 million images.
• I developed a vision-only provenance algorithm that reduced similarity-analysis runtime from months to minutes. At the time of the original project write-up, a manuscript describing the work was in preparation.

I tutored statistics and graded coursework for Econometrics I in the Master of Finance program.

I worked across machine learning applications and efficient data analysis for automated guided vehicles.

• I optimized the end-localizer program in ROS and deployed it to new products.
• I developed a computer vision system in OpenCV and PyTorch to determine whether two shelves were aligned.
• I built an end-to-end storage-status detection system covering data collection, labeling, model training, deployment, database design, communication with the warehouse control system, and a front-end interface. The system used Roboflow, Django, Redis, Celery, PyTorch, and ONNX to aggregate and update results in real time under constrained compute resources.