Jason Phillips, PhD

Across a year of heavy applied work in generative models, I implemented a wide range of diffusion-based techniques both in experiments and in production — from low-rank adaptations to ControlNets, IP Adapters, custom attention weighing and masking mechanisms (eg. in FlexAttention), image editing models & trajectory inversion, alignment with vision models, multimodal inputs, facial recognition integration, multi-region diffusion, custom classifier-free guidance variations with multiple controls, sampling techniques, etc. My projects bridged multiple backbone architectures, both convolutional and DiT.

I also spearheaded both research into multi-modal LLMs and their integration with image generation models, as well as the growing intersections between autoregression-based & diffusion-based generation (early versus late fusion, encoders and discrete or continuous patch-based embeddings, integrated versus separated attention modalities, and newer denoise techniques). I led regular paper sessions walking closely through the latest innovations and their implications.

I was fortunate to train at a large scale of GPUs using ray, and collaborated in the creation of a new foundational diffusion model with SoTA editing capabilities, for which I designed and implemented novel, advanced techniques for the creation of large-scale synthetic datasets.

Example project: Diffusion & Multimodal Generative Models

As a senior developer, principle developer, and finally Engineering Manager at Slite, I led and steered the core piece of the product—its advanced, fully collaborative text editor backed by operational transform. I participated at all levels of ideation and project execution, and worked often on low-level optimization, complex features, and even cooperating to design & launch a major NLP-based feature which is now a defining project of the company.

Slite is also a leader in the space of remote work, both as its internal way of operating (an international remote team), via its products which target remote collaboration, and in its positioning as a high profile player in the ongoing development & debate over remote work practices. I operated fully remote during all my years at Slite, even while managing a team with very distant international timezones.

Example project: Ask by Slite

Being the lead developer for this AI startup meant becoming an expert in the very latest industry trends in NLP, while also taking on the task—unmatched in the industry—of building a fast NLP pipeline to run entirely in the browser, through leveraging coordination with code compiled to Web Assembly. The entire stack relied on transformers, rust & C++ compilation to WASM, low-level language models like POS tagging and dependency parsing, ontology & language graphs (WordNet) modeled into vector spaces, and other techniques.

More: Browser-based NLP

Working for the Office of Institutional Research for over a decade, I was responsible for constructing numerous data-heavy applications (including the use of machine learning), integration layers, and dynamic presentations, either for the campus community, the administration, or for external bodies and accreditors. As leader of a small web-focused team, I hired and supervised junior developers and graduate students across many projects, spearheading all technical decisions and carrying out projects from initial idea through to stakeholder presentations and deployments.

Selected projects: Credential Mapping, Admissions Tracking & Visualizations, Unified Syllabus Editor, Dynamic Data Reporting, Survey Text Analysis, Accreditation Preparation, Committee Staffing, Tuition Remission

I reshaped an existing prototype application at this start-up by moving its entire data layer into a very large GraphQL schema, and built components and elements across its features, while leading other team members in instruction on best practices and testing design.

More: Brewery Software

• 2017 PhD in Educational Philosophy, Theory, & Foundations - The University of Alabama
• 2008 MA in Philosophy - Boston College
• 2005 BA in Philosophy (minors in German and Music) - The University of South Carolina

If you wonder how my doctoral work on John Dewey in educational theory is connected to contemporary problems in the space of artificial intelligence & LLMs, ask me—and prepare for a discussion of his early papers on neural responses to stimuli.

More general information on my academic background is available here.

Having conducted many technical interviews over the years—and knowing how difficult it is to grasp an applicant’s capabilities from a list of bullet points without at least hearing them walk through past projects—I constructed this site to offer background information on many areas of my work, even if it is necessarily far from comprehensive. Please explore the links above, but do not hesitate to contact me for further information. References available upon request.