Selected work

Proof you can inspect directly.

These are public engineering systems, not customer testimonials or private client case studies. Read the code, decisions, and constraints directly.

01 · AI systems

Odinn-Forge

Problem. AI assistants often hide memory, tool use, and decisions behind a chat window.

Work. A local-first assistant designed around inspectable memory, approved tools, and an auditable activity trail.

Evidence. Public source, tests, and documentation show the current implementation and its boundaries.

Open public repository

02 · Security engineering

GhostMCP

Problem. Security automation and agent tools are difficult to trust without strict authorization boundaries and auditable execution.

Work. A hardened Model Context Protocol server featuring per-engagement capability manifests, fail-closed proxy routing, and tamper-evident audit logging.

Evidence. Public source, tests, and documentation expose the security model, authorization controls, and runtime gates.

Open public repository

03 · Application security

security-middleware

Problem. Headers, CORS behavior, dependency checks, and security defaults are easy to scatter across an application.

Work. Security linting middleware for Node.js and Next.js that brings those checks and developer feedback into one inspectable path.

Evidence. Public source and test coverage show both the controls and the limits of what middleware can prove.

Open public repository