AI Website Operations System
I designed and developed a daily audit workflow that turns website findings into prioritized work, then checks whether completed fixes actually resolved the problem.
- AT A GLANCE
- Role
- Built the MCP integrations, audit workflow, Slack delivery, and tracker
- Timeline
- First working release in one week
- Stack
- React · TypeScript · Custom MCP servers · Anthropic API · Slack · Supabase · Vercel
- Outcome
- Daily audits across four sites; failed fixes reopen automatically
From finding to verified fix.
Test two outcomes from the workflow. This demonstration uses example data and sends no requests to client systems.
Broken internal link
A resource link interrupts the path to an installation guide.
/resources/installation-guideHTTP 404 · Not foundRun the workflow to see the recommendation and human review.
- 01DetectCheck link response
- 02PrioritizeAI recommends an action
- 03Review & resolveHuman completes the fix
- 04RecheckNext audit tests the destination
- 05ReportUpdate tracker and Slack
Choose an outcome. Follow the issue from detection through verification.
The audit summary will appear here.
The problem
Four regional websites needed consistent review across technical health, user experience, conversion opportunities, and SEO, GEO, and AEO. Findings also needed a place to become actionable work, with evidence, priorities, and follow-through.
I built a process that brings the findings into daily operations and connects a completed task to a fresh verification check.
What I owned
I created the custom MCP integrations, configured the scheduled audit workflow and Slack delivery, and designed and developed the internal tracker. AI supports interpretation, recommendations, and prioritization; I review findings and work through the remediation.
How the system works
Daily audits gather findings. AI interprets the evidence, recommends improvements, and assigns priorities.
Slack summaries surface findings and a priority breakdown before the workday.
The tracker groups findings with corrective actions, verification criteria, implementation notes, and progress states.
The next audit rechecks issues marked complete since the previous run. Failed checks reopen the issue and trigger a Slack flag.
AI with evidence
The workflow combines checks for code conflicts and broken links with recommendations for usability, conversion, and search visibility. A confirmed technical error and an improvement recommendation require different judgments. The tracker keeps supporting evidence and recommended actions visible so I can evaluate what to change.
AI helps interpret and prioritize the findings. Production changes remain a development and review step rather than an autonomous action by the audit system.
Verified follow-through
Marking a task complete does not end the process. Each daily run rechecks newly completed issues. A persistent problem automatically returns to the work queue and is flagged in Slack, making an unsuccessful fix visible instead of leaving it marked complete.
The interface supports regional and priority filters, remediation phases, blocked work, and implementation notes. These details make a broad audit easier to turn into focused development work.
The result
The system creates a consistent daily process for identifying issues across four sites, deciding what needs attention, tracking remediation, and verifying the result. It brings monitoring, AI interpretation, team communication, and a usable internal interface into one workflow.
This project demonstrates my ownership across integration development, AI workflow design, frontend implementation, and the operational follow-through required to maintain websites after launch.