AI Agent & Automation Pipelines

Production agent loops and execution harnesses running unattended: from a content pipeline that drafts, enriches and routes work for human approval, to the PRDs and technical specs behind them.

LLM APIs
RAG
n8n
Agent Orchestration
Teams Adaptive Cards
teams.microsoft.com/l/adaptive-card

Agent pipeline

DraftRAG enrichHuman reviewPublish

Content Automation Bot

New post ready for review

“3 changes to Irish VAT thresholds you need to know before Q3 filing…”

Source: RAG · 4 docsChannel: LinkedIn
ApproveRequest changes

Illustrative interface with fictitious sample data, shown in place of confidential client screens.

The problem

Recurring content and operational work (client comms, compliance updates, internal reports) was entirely manual, and every piece needed someone to draft it, check it against source material, and get sign-off before it went out.

What I built

An end-to-end content automation pipeline: n8n orchestrates the flow, an LLM API drafts and a RAG step grounds the draft in the practice's own reference documents, and the result is routed to a human for approval via Microsoft Teams Adaptive Cards before publishing.

More broadly, I own the AI delivery cycle from PRDs and technical specs through to implementation: agent loops, execution harnesses, and the tool integrations that let these pipelines run unattended in production rather than needing a human to babysit every step.

Impact

Recurring manual publishing effort dropped to near zero. A human still approves every piece, but drafting, sourcing and formatting no longer take their time.