Models & perception
Representation, memory, reasoning, and evaluation for AI systems.
Daedal develops perception models, knowledge tools, and software infrastructure.
Variable-rate video tokenization for longer, faster agent perception.
Local-first tools for querying documents and tabular data with citations back to source evidence.
Git-versioned specifications that stay readable to people and precise enough for coding agents.
A reproducible LLM research journal with pinned sources, measurements, evidence, and verdicts.
Representation, memory, reasoning, and evaluation for AI systems.
Retrieval and structured query systems that keep claims linked to source evidence.
Specifications and tooling for reproducible, agent-assisted software development.
Turn the broad ambition into a claim that can be disproved.
Make the smallest system that exposes the hard part.
Test on real workloads and preserve the evidence.
Promote what survives into reusable product and infrastructure.