Projects

doat.ai exists to make fragmented corporate data usable, by combining agentic retrieval with a navigable 3D view of an organization's knowledge graph.

Metis Stream exists because agent improvements are usually shipped on intuition; it requires a new skill to beat the previous baseline on replayed historical trajectories before it reaches production.

harness-injector exists to separate what an agent knows from what it can do, injecting retrieved knowledge into a session while exposing execution tools over MCP, so the context window stays lean.

code-gen-bridge exists to let external events trigger autonomous code generation, running the Claude Code CLI inside sandboxed workspaces with every step captured in a structured trace.

slack-block-kit-builder exists so that LLM agents can compose rich interactive Slack messages without holding Slack tokens or requiring network access.

Context Engineering Lab exists to give junior developers a hands-on way to compare RAG, prompt engineering, LLM parameters, agent tool-calling, and GraphRAG side by side.

low-poly-generator exists to produce low-poly animations and wireframe visuals for the web, the same generative-geometry approach used in this site's hero.