All work
LLM automation that gave a team back hours every week
An n8n + LLM pipeline — built for a client (name withheld for confidentiality) — that took a recurring manual triage-and-draft task off a person's plate.
Focus:
- n8n
- LLM automation
Type:
Client project — confidential
Discipline:
AI application engineering
Tags:
- #n8n
- #LLMautomation
- #Structuredoutput
- #Confidentialclient
Note: client identities and full system architecture are withheld for confidentiality — enough is shared here to show the shape and substance of the work.
Problem
For a client (name withheld), a recurring task — read incoming items, classify them, and draft a structured response — was eating hours of someone's week and easy to fall behind on.
Approach
- Mapped the manual steps and identified which were judgement-light enough to automate safely.
- Built an n8n pipeline that ingests items, calls an LLM for classification and drafting, and returns structured output.
- Added confidence thresholds so low-confidence items are flagged for a human instead of auto-actioned.
- Set up alerting and a simple review queue so nothing fails silently.
Result
- The bulk of the recurring work now runs unattended on a schedule.
- A person's weekly time on the task dropped substantially, freeing them for higher-value work.
- The confidence gate kept quality high by keeping a human on the genuinely ambiguous cases.
Architecture
