What is parralax?

Parallax is a hackathon project that tries to solve a specific trust problem in AI automation: business software can record that a task was completed, but nothing checks whether the physical world actually matches that record. Parallax is described as a verification layer that uses a camera to confirm reality before an automated process is allowed to continue. The project works alongside a separate AI orchestrator called Hermes Agent, which decides what steps to take next in a workflow. Parallax itself does not make those decisions. Instead, a photo is taken of the real world situation, such as a packed shipping box or a finished repair job, and a vision AI model reads out what it actually sees in that photo. That observation is then checked against a fixed set of rules written in regular code, not by another AI vote, to decide whether the work is genuinely complete. If something is missing or wrong, the system holds the process and generates a specific list of what needs fixing. Once corrected evidence is provided, it checks again. Two example scenarios are given in the README: a warehouse packing station that verifies a box contains the right mug, cable, and warranty card before it ships, and a repair technician job that checks a safety seal and label were actually installed before the visit is marked closed. The project also supports defining a brand new custom scenario just by uploading reference photos, without writing any extra code. Every decision the system makes is recorded to a tamper evident, hash chained log so it can later be independently verified. The project is built with TypeScript, React, and Express, and can run entirely in a local demo mode using saved sample data with no external API key needed, or connect to a real vision model through Hugging Face or NVIDIA's cloud service for live use.