Shikshak-Saathi

Pitch Deck:https://drive.google.com/file/d/1ncPjwrtKKEv9C_TEA6tuvEfd4DcsjsIp/view?usp=sharing

Problem: Teachers spend about only 19% of their time teaching. Over 1,04,125 Indian schools run with a single teacher, and 42% of students study in Hindi medium. Most AI tools today are English-first, not tied to NCERT, often paid, and assume the classroom has a projector.

The solution: A teacher picks a class, subject and chapter. Shikshak Saathi then produces:

- Learning objectives: each tagged with its Bloom's taxonomy level, NCERT page references, the prior knowledge students need, key terms, and the mistakes students commonly make.
- Timed lesson plan: split into starter, explanation, activity, practice and exit, with a script for the teacher and a list of materials. The timing is adjusted to match the actual period length.
- Low-resource and multi-grade support: chalkboard layouts, and separate tasks for each grade when one teacher handles several grades in one room.
- Worksheet with answer key: generated together so the answers always match the questions.
- Formative assessments: exit quiz, starter quiz and chapter test. In the exit quiz, every wrong option is linked to a specific student misconception.
- Parent note: homework summary that can be shared on WhatsApp.
- Export: as PDF or Word, for printing.

After class (closing the loop):
- The teacher enters quiz results, or just photographs students' paper answer sheets. The AI reads the circled options and handwritten answers.
- Misconception analytics list the class's biggest mistakes, show accuracy per question, and show how well the class has mastered each objective.
- It then suggests a 5-minute fix for the next day's lesson.
- Voice input: teachers can speak notes and instructions instead of typing (Sarvam AI speech-to-text and text-to-speech).

What makes it reliable:
- Step-by-step generation: each section is generated one at a time, in order, and can be regenerated or edited by the teacher.
- Automatic checker: 12 fixed rules (no AI involved) check the output and fix problems like wrong timings or incorrect worksheet totals.
- Backup AI providers: it switches between Gemini keys when one is rate-limited, falls back to OpenRouter, and has a cached demo mode, so a live demo never breaks.

Tech stack: Next.js 16 (App Router, TypeScript), Tailwind v4 + shadcn/ui, Prisma 7 + PostgreSQL (Neon), Better Auth, Google Gemini (including its image-reading ability for scanned sheets) with OpenRouter as backup, Sarvam AI for voice, installable as a mobile-friendly web app (PWA), hosted on Vercel.

Built at: Hack-e-Awadh 2026 (problem statement PS-01: Teacher Lesson Planning Agent) by Team NawabiCoders, Lucknow.