Case studies
Reference implementations in production.
Systems vuihoc.ai designed, built and operates. Each case study documents the problem, the architecture, the controls and the measured result, so you can judge the method before engaging us.
-
Enterprise knowledge assistant
Semantic search across 42,000+ documents in isolated workspaces, available in Claude on desktop, web and mobile, with cited sources.
42,000+ documentsWorkspace isolationCited answers -
Evaluation framework
Stratified test sets, multi-provider benchmarks and calibrated LLM judges that replace intuition in model and prompt decisions.
192-case test set6 models benchmarkedCalibrated judge -
Knowledge integrity audit
A three-layer audit and four-tier schedule that detects outdated facts and cross-source contradictions in a knowledge base.
25+ drifted facts found on day oneAccuracy 80% to 99% -
Priority inbox triage
An agent that ranks email threads P0 to P2, verifies against project tools, and surfaces only items that require action.
~90 to ~8 min a day568 to 7 priority itemsHuman approval -
MCP connector hub
One governed connector layer giving Claude access to Google Workspace, Slack, GitHub, Jira and Linear across three isolated workspaces.
150+ tools3 isolated tenantsOAuth and passkeys -
Vendor documentation retrieval
A separate authoritative index of AI vendor documentation, with routing that prefers it for pricing, specification and version questions.
Authoritative-source routingSeparate index -
AI coding governance control plane
Per-developer token visibility, daily caps, spend alerts, a central off switch and CI cost reporting for a coding-agent rollout.
7 controlsCentral off switchPer-developer KPIs -
Agent-readable diagram engine
A text-to-diagram language designed for LLM authors, with deterministic layout and validator messages a model can correct in one pass.
37 layout rulesOne-pass repair -
Website audit and rebuild
This website, assessed against Apple HIG and WCAG with measured findings, then rebuilt from a personal blog into a services site.
HIG and WCAG 2.2Measured findings
Start here
Have a problem AI might solve?
Describe the outcome you need. Within two working days you receive a written view on whether AI is the right tool, the risks to manage, and a proposed first step.