SystemAudit MCP server. Ask Claude or Cursor "is this codebase safe?" and get a straight answer in seconds, from a tool I built that is listed in Claude's connector directory.

_{Under the hood: remote MCP server, three read-only tools. No model call: every figure comes from a deterministic scan, with files read stated.}

DocsFlow, now SureCiteAI. A private AI assistant for a company's own documents that shows the source behind every answer. Tested on 297 questions without inventing a single source.

_{Under the hood: hybrid retrieval with rank fusion, four tenant-isolation layers, four-model failover, abstention when evidence is weak. 0 hallucinated citations in 297 benchmark cases.}

AI NeuroSignal. Twenty AI agents with different strategies read the markets, and the system learns which ones to trust by checking every call against what really happened. Over 38,000 predictions tracked live.

_{Under the hood: parallel LLM calls with schema validation; deterministic voting that abstains on disagreement; 38,319 predictions scored against outcomes.}

I studied 53 apps built with AI tools like Lovable, Cursor and Claude Code. One in five had their passwords and keys committed to the code, and nearly three in four had no automatic security checks at all. AI makes apps fast to build, not safe to launch; this shows exactly where they break, with a checklist to fix it.

_{Under the hood: deterministic scan with no AI model in the loop, every credential finding checked by hand, anonymised dataset published. Citable via DOI.}

Cairn, three Claude Code plugins. AI coding assistants often say "fixed" when they haven't. These plugins check the AI's work, so a team can trust what it reports. Free and open source.

_{Under the hood: a script locates, the model judges. Every check has a self-test that plants a defect, run in CI. Results published, including a null.}

SystemAudit. Paste a GitHub link and get a plain-English health report on any codebase in under three minutes: the kind of review that usually takes a consultant weeks.

_{Under the hood: open-source scanner engine (dependencies, import graph, health scoring) feeding a multi-pass AI layer. Findings carry file-and-line evidence.}