Weekly trend-to-schedule content pipeline that repurposes its own winners.

A weekly cron pulls Google Trends and Twitter recent-search data, GPT-4 builds a content calendar and per-platform copy, and DALL-E generates images. Output saves to Sheets and Airtable and schedules through Buffer, with a Slack report each run. Seven days later it analyzes engagement and repurposes the top performers.

RAG support triage with grounded drafts and timeout escalation.

Normalizes tickets from Zendesk, Intercom, email and webhooks, then classifies intent, urgency and sentiment with OpenAI. It retrieves relevant articles from a Pinecone vector knowledge base via embeddings and grounds the drafted reply in what it actually retrieved, then routes to the right team. A two-hour no-reply branch escalates anything left unanswered.

Alert to diagnosis to gated remediation, with an automatic post-mortem.

Normalizes alerts from Datadog, Grafana and scheduled health checks into a single shape, then GPT-4 diagnoses root cause and sets confidence and business-impact ratings. Conditional branches drive the response: AWS auto-scaling, an ECS service restart, or a GitHub rollback, each gated rather than fired blindly. Notifications fan out to Slack, PagerDuty and email, and the run writes a post-mortem to Notion and GitHub.

Throttled outreach with LLM reply classification, compliance guards and a remote kill switch.

A set of coordinated n8n workflows over a Convex backend and Close CRM. Leads arrive by webhook into Convex. A weekday batch builds a personalized text-first opener per lead and sends it through the CRM API on a 20-second throttle for rate-limit safety. Replies return through a webhook that runs a STOP and opt-out compliance guard before anything else, then an LLM intent and sentiment classifier, then a routing call that flags interested leads and fires the downstream build-and-send. An operations layer runs alongside: a webhook kill switch that flips a runtime flag in the backend, an hourly health check that alerts Slack when the system is unhealthy, and a dedup sweep that quarantines leads already sitting in the CRM.

Daily property sourcing, LLM scoring and CRM delivery for a US real-estate investor.

A four-stage n8n system that runs every day for an investment client. It scrapes listing sources across a set of target ZIP codes, hashes each address to deduplicate, and idempotently upserts clean records into Postgres. Leads that survive the first filter get a detail scrape, then enrichment from Gemini and the RentCast valuation API, which together produce a deal score and a Maximum Allowable Offer. Top leads are pushed into Close CRM with scores and push history written back to Postgres, plus a Slack digest and per-lead alerts. A separate webhook handles on-demand comping: post a subject property and get back a geocoded valuation and MAO in the response. It delivers around 25 qualified leads a day.