27 Mar 2026 7:18 PM MST

Watch on YouTube:YouTube

Robin Allenson leads our 6th session of Coding with Claude.

Watch our group:

1. Ingest our product requirements into Claude.
2. Decide how our tool should work and what level of detail we want in each feature.

Wondering how to go from idea to app using Claude Code? This is the first step.

What we did:

• Named the tool — landed on Traffic Diagnosis (handles both spikes and drops)
• Initialized a new git repo and pasted the product requirements into Claude Code
• Worked back and forth with Claude to scope out the full feature set

What we decided:

• Architecture: Local install only (not centrally hosted) — data privacy concerns around storing client GSC/GA4 data
• User: Single agency user, managing multiple clients, starting with one Google account
• Detection: Start with A + B — identify what dropped/spiked (URLs + associated queries); skip the AI "why" diagnosis (option C) for now
• Anomaly logic: Statistical deviation from normal (percentage-based), so it works for both small and large traffic sites; no fixed thresholds
• Alerts: Reactive first — proactive/scheduled monitoring is v2 (requires a server running 24/7)
• Branded queries: Check if the GSC API already has a branded filter before building our own
• Query grouping: Use vector embeddings — FastText floated as the cheap/fast/good-enough option
• Reporting: Charts, tables, and diagnosis flags exported to Google Sheets for v1; branded PDF reports are v2
• UI: Build a web app (not just command line), something like a Radiant-style dashboard

What's next:

• Robin will wrap up the scoping conversation with Claude and give it the remaining answers
• Create and share a GitHub repo so the community can start contributing