LoomDB
A graph-based memory substrate for AI agents, inspired by temporal decay and human memory models.
Built to reduce context bloat and make long-running agents more relevant over time.
RustGraph MemoryAI Agents
AI Engineer · Agent Systems Architect · São Paulo, Brazil
I build AI systems that remember better, retrieve smarter, and stay useful over time.
From agent memory architectures to low-resource language models, I'm interested in systems that carry context, adapt, and do real work outside the demo.
About
Jorge Guberte
I'm a software engineer focused on AI systems, agent architectures, and interfaces for working with context over time. I'm especially interested in memory models, retrieval, orchestration, and the weird edge where software starts behaving more like a cognitive system than a static tool.
Read more about my work →Now
- Building LoomDB, an experimental memory substrate for AI agents.
- Exploring long-term memory, context engineering, and MCP workflows.
- Running and testing LLM inference systems on Modal.
- Designing AI products and agent workflows at Hub Esfera.
Selected Work
Projects with a strong point of view.
Multiverse
A state management library with Git-style branching semantics.
Useful for applications that need forkable state, experimentation, and timeline-safe UX patterns.
TypeScriptStateReact
Ayvu-Talian
A decoder-only transformer trained for Talian, a minority language spoken in southern Brazil.
An exploration of low-resource NLP, language preservation, and model-building from scratch.
NLPPyTorchLow-resource LMs
Labs
A playground for experiments, interfaces, and useful weirdness.
Static, fast, and intentionally scrappy. This is where prototypes can exist before they deserve a product page.
Writing
Notes from the builder side of AI.
Notes on memory, context, infrastructure, and the realities of building AI systems outside toy demos.