Dallas Self
Creative Technologist | VFX Artist | Machine Learning Engineer
I build visual and technical systems where art, VFX, and machine learning overlap. My work blends production-minded problem solving with creative tools, from compositing and real-time visuals to AI-driven experiments and applied ML projects.
Portfolio
RAGCraft: Episodic Memory for RAG Agents in Dynamic Environments
Machine Learning / RAG / Agent Memory / Minecraft Testbed
About
I’m Dallas Self, a creative technologist with a background in VFX, computer animation, custom illustration, and machine learning. I build visual work and technical systems, usually with one foot in art and the other in problem solving.Before getting deeper into machine learning and technical workflows, I spent years making visual work through illustration, compositing, and production. That background still shapes how I build things. I care about how something looks, how it works, and that it is stable, scalable and dependable when needed.Right now, I’m focused on applied machine learning, AI-assisted VFX workflows, real-time tools, and creative systems that help artists' effeciency but lets them stay in control.
Contact
Want to talk about a project, role, collaboration, or strange technical art problem? Send me a message.I’m open to opportunities in machine learning, creative technology, VFX, AI-assisted workflows, and technical art.
RAGCraft Case Study
Machine Learning / RAG / Agent Memory / Minecraft Testbed
The Problem
Most RAG systems retrieve information, but they do not meaningfully learn from their own repeated experience. RAGCraft explores whether an agent can improve across repeated tasks by storing compact experiential memories based on goals, mistakes, corrections, and outcomes.My Approach
I built a Minecraft-based test environment with repeatable tasks, including lever puzzles, key retrieval, and maze navigation. The agent was tested under two conditions: raw retrieval and distilled memory. The distilled memory system compressed each attempt into structured experience units that could be retrieved and reused in later runs.What I Built
A scenario runner, memory pipeline, retrieval system, experimental logging, evaluation scripts, and visual test environments. The system tracks success rate, completion time, attempts, and resource usage across repeated runs.
Results
Distilled episodic memory improved task completion and reduced repeated errors across multiple scenarios, especially after the agent had accumulated enough experience to reuse prior corrections.
Memory Diagram and Tech Used
Core stack: JavaScript, Node.js, Python, JSON, Git/GitHub
AI/ML: RAG, embeddings, vector retrieval, episodic memory modeling
Environment: Minecraft, PaperMC, Mineflayer
Evaluation: experimental logging, scenario runners, data analysis, visualization, agent behavior metrics