Case study
AI automation / Deployed workflow
AI Request Router
A Render-deployed n8n workflow that classifies inbound requests, routes actionable signals, and preserves uncertain cases for human review.
- Role
- Creator · Workflow architecture
- Year
- 2026
- State
- Deployed on Render
System / routing
Scroll to examine
Project snapshot
The condition
Challenge
Inbound sales, support, spam, and general requests arrive as unstructured text. Automating triage is only useful if malformed model output and uncertain classifications remain visible instead of disappearing into a silent failure path.
The response
Architecture
A nine-node n8n workflow sends each message to Llama 3.3 through Groq, parses a strict category-priority-summary contract, then routes valid classifications to Slack and Google Sheets. Invalid output follows an explicit fallback branch that logs the raw response for manual triage.
What shaped it
Key decisions
Constrained model contract
The classifier requests strict JSON, then a defensive code step normalizes and validates the response before routing it onward.
Failure is a first-class route
Malformed output is sent to a dedicated Slack alert and raw spreadsheet log so a person can resolve it with the original evidence intact.
Reproducible self-hosting
Pinned n8n and PostgreSQL images, health checks, encrypted credential storage, and environment-separated secrets keep the local system portable.
How it feels
Experience
A dependency-free browser client submits a message and returns its category, priority, and summary. The team receives a Slack signal and a spreadsheet audit record, while malformed classifications remain recoverable through the fallback route.
What resolved
Outcome
A live browser demo backed by a Render-deployed n8n service and PostgreSQL, with an importable workflow and reproducible Docker Compose stack for running the complete classification and recovery path.
- Live browser demo
- 3-resource Render stack
System composition
Technology, deliberately chosen.
- n8n 1.70
- PostgreSQL 16
- Docker Compose
- Groq
- Llama 3.3
- Slack