Redbook

Claude API · TypeScript · RAG (pgvector) · Vercel · Notion API

• The customer support team was overwhelmed by inbound questions. The answers lived in Notion and support could access it, but hand-searching the docs for every inquiry was slow.
• Customer-facing comms (release notes, migrations) were a manual handoff from engineering to marketing every two weeks, with predictable drift.
• Onboarding a new customer meant 3 to 4 hours of manual data wrangling from their previous software.

• Visited feed yards in person. Shadowed cowboys, owners, and managers using the product to turn field observations into prioritized engineering work.
• Owned end-to-end integration of live cattle futures market data. That meant data ingestion, the modeling layer, and the UX that surfaces real-time P&L predictions to customers.
• Built a Claude-powered TypeScript agent on Vercel, with a RAG pipeline (pgvector + OpenAI embeddings, incremental sync via Notion last-edited-time) connected to the OpenPhone API to auto-draft grounded responses to inbound customer questions.
• Built a second agent that pulls merge commits from the repo every two weeks and generates polished product-update emails. That replaced the manual engineering-to-marketing handoff.
• Built an automated migration pipeline from competitor software.

• Live market-data pipeline powering real-time P&L predictions in-product.
• Notion-grounded internal support agent that auto-drafts answers to customer inquiries via OpenPhone.
• Automated bi-weekly product-update email pipeline.
• Automated migration pipeline that turns 3–4 hours of manual setup into minutes for every new customer (10+ onboarded so far).

• Expand the agent toolset so it can take action, not just answer.
• Deeper instrumentation on which retrieved docs actually drive resolved tickets.