hijohnnylin/neuronpedia — reverse-engineered prompt

Build me an open source AI interpretability platform like Neuronpedia. I want a web app where researchers can browse model features and neurons, search through explanations, inspect activations, view dashboards, compare similar features, and save useful lists. It should support uploads and exports so people can add their own data and share results.

Include a simple API behind the app, plus pages for circuits or graph views, activation testing, steering experiments, automatic explanations, scoring, benchmarks, embeddings, probes, and visual maps when data is available. Make the local version usable with an empty database, with an admin flow to import sources, activations, explanations, and related data later.

Please set it up as a TypeScript project that can run locally for web development, with clear environment setup and commands. Keep the interface clean and research focused, not flashy. If you need exact implementation details for interpretability concepts or local inference setup, look up current docs online.