15 PROJECTS · 4 CATEGORIES · 2023—2026
Selected Work
METHOD DEPTH
From model training to interface craft.
A second axis over the same work — not where a project lives but what kind of technique it is. Top to bottom: building and adapting model weights, then working in latent space, orchestrating agents, running models at inference, and finally pure interface and geometry craft. Tags are evidence-based; a hollow marker is a secondary method.
From-scratch training
Designed and trained a model's weights from the ground up.
Fine-tuning
Adapted a pretrained model (LoRA, unfreezing, differentiated LRs).
Embeddings / latent space
Builds on learned representations — projecting, navigating latent space.
Agentic / orchestration
Agent loops, tool-calling, LLM pipelines and self-verification.
CV / model inference
Runs vision / generative models at inference — real-time, on-device.
Interface / interaction
Interaction, dataviz, geometry and design craft — little to no ML.
No projects match. Reset filters.