Tuff Shed Scripts

Alongside the main Tuff Shed platform, the team built a set of Python scripts for internal automation. Two projects stand out.

Semantic Document Classifier

The most interesting tool is a file organization utility that uses the sentence-transformers library (all-MiniLM-L6-v2) to semantically classify documents into configurable categories. It handles PDFs, Word documents, Excel files, and images via OCR with Tesseract, and logs every operation for transparency.

It’s a small project, but it captures a practical approach to automation — take a repetitive task, make it smart enough to handle edge cases, and keep it auditable. The tool eliminated hours of manual file sorting across the team.

AI Agent Project (Alpha)

This repo is also the foundation for an alpha-stage AI agent project currently in development. The goal is to build a Claude-powered AI agent that can eventually take on the responsibilities of a product manager — using the two years of context, documentation, API collections, and domain knowledge gathered throughout the Tuff Shed engagement.

The first step is organizing and semantically analyzing all of that accumulated content — Postman collections for Nextworld APIs, test data objects from NetSuite, domain-specific JSON schemas — so that an agent has enough structured context to reason about the product and make informed decisions. It’s early-stage work, but it represents an investment in AI-augmented product management and a belief that the best way to hand off institutional knowledge is to make it machine-readable.