In Simple terms: a Searchable Database that stores MEANING – Not Keywords. A production retrieval pipeline — Telegram capture, LLM-based extraction, vector embeddings, and an MCP server sitting in front of Supabase/pgvector — queryable in plain language from Claude, ChatGPT, or any MCP-compatible client. Every proof point in this document was run live against the […]
A production pipeline I designed, built, and have run solo for over 12 months. One Telegram message — voice or text, English or Spanish — routes through Whisper and GPT with a context-injected system prompt, and lands as a structured, owned, deadlined row on my master board. Prompt engineering, strict JSON schema, multi-branch Make.com orchestration.
LLM Workflow Automation — One Capture Point, Zero Dropped Tasks Read More »
A Case Study: Zero-entry expense capture — an agentic workflow built in Google Workspace Studio + Gemini ABSTRACT Cara Mía Theatre runs half a dozen programs at once — Latinidades Festival , shows, summer camp, educational residencies, and more — and every one of them generates its own steady stream of receipts and invoices. This
Fixing Bookkeeping with AI Read More »
The Rat Brain of Project Management: A Simple System for Real Execution If you had to teach project management to something that could only understand the basics—no abstractions, no complexity—what would remain? Three things: Why Simplification Wins The concept is loosely inspired by behaviorist thinking, especially researchers like B. F. Skinner, who explored how behavior
The Rat Brain of Project Management Read More »
