jason.lunder@pangeon.com | github | linkedin | Resume | CV | blog

Founding Software/ML Engineer at Intellipat Inc.
Master’s Student at Eastern Washington University

I build AI and software systems for patents and intellectual property law. I graduated from Gonzaga University with a BS in Computer Science and a BS in Mathematics in 2023, and I am currently pursuing an MS in Computer Science at Eastern Washington University, defending my thesis Winter 2026.

Research

My research focuses on natural language processing, with an interest in structure-aware language representation, especially tree-based models that leverage linguistic structure. I am exploring how combining structured and unstructured linguistic representation can improve efficiency, interpretability, and safety in AI systems.

Tree Matching Networks

Adapting graph neural network architecture for natural language inference using dependency parse trees. Explicit structural encoding outperforms transformer baselines at matched parameter scales. My master’s thesis (Winter 2026) investigates self-attention aggregation to address known scaling challenges in structure-based approaches.

Publications and Preprints

Lunder, J. (2025). Tree Matching Networks for Natural Language Inference: Parameter-Efficient Semantic Understanding via Dependency Parse Trees. arXiv preprint arXiv:2512.00204. https://arxiv.org/abs/2512.00204

Experience

Founding Software/ML Engineer at Intellipat Inc. / Pangeon Corp.
Team Lead at Parsimony
Machine Learning / Data Science Intern at Gestalt Diagnostics
Research Assistant at the Gonzaga University Center for Complex Systems

Projects

Tether, An open-source agentic AI task management and planning application. Client/server architecture with agentic AI integration for intelligent task orchestration. Licensed under AGPL-3.0.

I build models and infrastructure for machine learning systems, including a system for determining if an invention is novel in the context of all patents, a system of various CV models for assisting pathologists diagnosing cancer, and a novel lemma dependency tree based machine translation model.

I led the Gonzaga Robotics CS and ML team in designing a control system for a mars rover. I contributed to the 2022 ACROBAT Grand Challenge winning image registration model for histology images. I developed an RPA based system for automating IT processes at Lynden Inc.

Education

MS Computer Science - Eastern Washington University (2024-2026)
BS Computer Science & Mathematics - Gonzaga University (2019-2023)

Tree Matching Networks: What If We Gave Neural Networks the Parse Tree Instead?

When we feed text to a transformer, the model has to figure out the structure of language entirely from data. Subject-verb relationships, modifier scope, negation boundaries: all of it must be learned implicitly from millions (or billions) of examples. But what if we just… told the model the structure upfront?

That’s the core idea behind Tree Matching Networks (TMN), my recent research exploring whether dependency parse trees can serve as an efficient structural prior for natural language inference. The short version: a graph neural network operating on parse trees achieved 75.20% accuracy on the SNLI benchmark versus 35.38% for a BERT baseline of comparable size, both trained from scratch on the same data and the same hardware.

By Jason Lunder

read more