VidyaTid

VidyaTid is a preparation platform for NEET, India's medical entrance exam. It began in September 2025 as a three-person hackathon entry, a tutor for JEE and NEET that answered from NCERT textbooks. Between January and September 2026 it was rebuilt as a NEET-focused product: a tutor grounded in NCERT material, a bank of past questions in English and Hindi, a timed mock-test engine, rank and paper prediction, streaks, and generated explainer videos.

The data. The question bank came out of source PDFs full of damage: entries that mixed two languages, null bytes, junk left by conversion, and mathematical symbols turned into the wrong glyphs. Exam preparation is unforgiving about wrong answers, so every repair had to be checkable and reversible. Answers were remapped only when at least three of four options matched, and every changed row kept a backup.

A Flask backend of about 67,000 lines with 14 feature modules and a set of agents: an orchestrator, and agents for explainer videos, rank prediction, mock tests, weekly tests, paper prediction and images. It runs on PostgreSQL 16 with pgvector (moved from SQLite in August 2026), ChromaDB and Redis, with Docker, nginx, metrics, error reporting, CI workflows and runbooks around it. The front end is Next.js 16: marketing pages, sign-in, an onboarding wizard, a test engine with a deadline clock and autosave, a dashboard and an admin console. The tutor calls Gemini and Cloudflare Workers AI; a local Qwen2.5-7B model writes questions for the prediction and mock-test agents.

What can be counted is the data. The publishable bank holds 39,385 questions: 18,100 physics, 15,754 chemistry, 3,900 biology, 296 mixed and 1,335 in Hindi. Cleaning took mixed-language entries from 2,236 to 0 and null-byte entries from 749 to 0, and the project's status file records 340,692 glyphs repaired across 10,995 questions. "Cleaned" means cleaned and counted against a checksum ledger; it is not a check that every answer is right. And the part that matters most to a student has not been measured at all: there is no figure for tutor accuracy, retrieval quality, or how well the rank and paper predictions hold up.

Not deployed. The public site at vidyatid.in is a waitlist page, not the product. Some dashboard views still show sample data, the early goal of running fully offline was not reached (the tutor depends on cloud models), and the licensing of the question sources has to be settled before any public launch. I paused the project in September 2026; the last code change on file is from 6 September. It stays here as what it is: about three quarters built, then paused.