Zidd.

Zidd is an AI English coach built for learners in India, and it runs two products on one engine. School teaches NCERT-aligned English for Classes 2 to 5 across 47 authored chapters, with a placement diagnostic, teaching sessions, recaps, and reports a parent can read. Career runs communication drills, mock interviews and group-discussion simulations for job-seekers. Learners answer in their own words, with explanations in English, Hinglish or Hindi. Every answer goes through a LangGraph decision graph that scores it and picks the next teaching move: teach a new idea, bridge down when the learner is lost, strengthen, challenge once they have it, or redirect when the answer is fluent but off the question. Every loop is capped so nobody gets drilled forever, and progression runs on a streak of clean passes rather than a moving average, because an average that settles at 0.8 never crosses a strict 0.8 gate and soft-locks a learner who is getting it right. The reply streams token by token over SSE. Underneath, it is built to stay correct on a free tier: boot fails closed if the secret, database or schema is wrong, open-answer grading runs through a Postgres job queue that resumes after a crash, session starts and credit charges carry idempotency keys so a retry never double-charges, and a pool of Gemini keys rotates on rate limits. Reports only state progress the saved evidence supports, so an empty report says so instead of guessing.