Traditional general-purpose AI systems lack curriculum context and grading scoping, often offering overly complex or out-of-scope guidance for K-10 primary and secondary school tutoring. VidyaAI solves this by enforcing curriculum boundaries and NCERT references.
Built a specialized triple-learning mode (Friendly Chat, Structured Study Notes, and Audio Overview) that dynamically scopes prompt generation. Offloaded retrieval grounding to browser-side PDF text extraction using pdfjs-dist, eliminating heavy server-side databases.
Developed a rich React and TypeScript client integrated with the high-throughput Groq LLaMA 3.3 70B API. Configured a dual-channel text-to-speech engine using Google Gemini 2.0 (oral script translation) and ElevenLabs (TTS voice synthesis) with browser-native Web Speech player fallback. Stored session summaries inside Postgres with Row-Level Security (RLS).
User Select → CBSE System Prompt Construct → Chapter File Upload (pdfjs-dist parse) → Groq LLaMA 3.3 70B API (SSE stream) → RLS Database Sync (Supabase PostgreSQL) → Gemini script gen → ElevenLabs audio compilation.
React & TypeScript
Modern component architecture and strong types ensuring high code quality
Groq API
High-throughput LLaMA 3.3 70B streaming for real-time tutoring chat
Google Gemini
Generates conversational, mathematical-formula-free spoken scripts for synthesis
ElevenLabs
High-fidelity audio synthesis providing natural educator spoken voice streams
pdfjs-dist
Performs zero-cost client-side text scanning and textbook file parsing in-browser
Supabase
Managed PostgreSQL database, Row-Level Security logs, and secure user Auth flow
Tailwind CSS
Sleek responsive design integrated with accessible Shadcn component layer
