Selected work

Proof that the system got clearer.

A small set of public work, presented with the problem, the intervention, and the operational consideration still in view.

01 · Security engineering

Security middleware

Problem. Application security controls were too easy to treat as scattered configuration. The useful work was to make headers, CORS behavior, dependency posture, and security defaults inspectable in one place.

Work. TypeScript middleware and documented checks that can be reviewed, tested, and carried into a real application.

Evidence. Public source, test coverage, and a clear limit on what middleware can prove by itself.

Open public repository

02 · AI automation

GhostMCP

Problem. Agent and tool workflows become dangerous when authority is implicit, inputs are untrusted, and an operator cannot see what happened.

Work. A policy-guarded security workflow in Python that keeps tool access, approval, and evidence explicit.

Evidence. Public implementation and documentation around the boundaries the system does—and does not—enforce.

Open public repository

03 · Full-stack delivery

Product delivery

Problem. A product can look finished while its deployment, observability, or handoff remains an afterthought.

Work. Interfaces, APIs, data flow, tests, and release documentation treated as one system rather than separate tickets.

Evidence. The deliverable is a working path from user action to persisted result, with the operating assumptions written down.

Representative engagement pattern.

04 · Cloud & DevOps

Infrastructure review

Linux hosts, service boundaries, deployment paths, and monitoring should be understandable before they become an incident. This lane covers the practical review that connects application behavior to the environment carrying it.

Representative engagement pattern.

05 · Data & analytics

Agent evaluation

Automation needs a way to be judged. Evaluation work turns a vague sense that an agent is “usually fine” into testable cases, failure categories, and an operator-visible record.

Representative engagement pattern.

06 · Open source

The rest of the lab stays inspectable.

The archive contains experiments, tools, agent runtimes, and security research. They are useful evidence of how BlueDot thinks, but they are not presented as client outcomes.

Browse the public archive