Radhikaag2006/urjakavach ? reverse-engineered prompt

Build me a secure internal AI workbench for a refinery team that runs completely on our own network and keeps all data private.

I want users to sign up and log in with Email, GitHub ID, or Phone Number plus a password, with hashed passwords, encrypted local user storage, and separate private chat history for each user. After login, they should be able to edit their profile details like name, profession, and country from a sidebar. The app should have a polished dark themed interface with smooth animated login screens and a modern feel.

The main product should support two workflows. First, a scanned inspection report should be read with OCR, checked against our own reference documents, and turned into a downloadable approval note in .docx format. Second, if a user gives a plain English coding request, the app should generate Python and run it in a safe isolated sandbox to verify it works before showing it.

Please include an audit log for everything important, make it work offline or with local models, and if anything is missing, use sensible stubs so the app still runs.