Disclosures

Required by the Suvidha AI Virtual Hackathon. Pass/fail, not scored.

AI tools

• Hermes Agent (Grok 4.6 via xAI) — coding assistant used to write this repository during the build window.
• Next.js, React, Tailwind CSS — open-source frameworks, not generative models.
• Groq (`llama-3.1-8b-instant`) or Featherless — optional runtime on a Node host only. The key stays in server env. GitHub Pages strips /api/diagnose. /demo does not call a model. Calls stay off until GROQ_ARMED=1.

Datasets

• None. Lena’s worksheet, blockers, and checks were written for this prototype. No scraped homework banks, no student data, no fine-tunes. Lena is a composite informed by being the user.

Built vs mocked

• Fully built, no model: `/` (also `/demo`, `/lena`) is the judge path — Lena’s worksheet, apply-to-problem-2, write the next line.
• Fully built, no model: `/open` is the tool. Load Algebra 2, Physics, or Calculus; paste; or upload a photo / PDF / Word file. Reading stays in the browser.
• `/api/diagnose` exists in source for an optional hosted model. It is stripped on GitHub Pages and is not required.
• Not in scope: chat tutor, gradebook, LMS, solving the remaining problems.