VIDYA EDUCATIONAL LLM11 INDIAN LANGUAGES
An open-source NCERT-focused educational AI companion powered by a fine-tuned 1.7B model (vedantjadhav701/edu-qwen-1.7b-merged).
Multilingual Educational Benchmark (v1.0)
Vidya 1.7B was benchmarked using the Vidya Multilingual Educational Evaluation Suite across 64 evaluation questions, 8 writing systems, and 4 STEM domains (Mathematics, Physics, Chemistry, Biology).
BENCHMARK PERFORMANCE SUMMARY
"Vidya achieves an overall accuracy of 93.3% (9.33 / 10) across NCERT STEM evaluation datasets, with near-perfect chemical equation solving (99.4%)."
Vidya Core Capabilities
Every feature is built specifically for Indian curriculum requirements, zero translation fallbacks, and mathematical precision.
11 Supported Indian Languages
Native fluency in English, Hindi, Marathi, Tamil, Telugu, Bengali, Gujarati, Kannada, Malayalam, Punjabi, and Maithili.
Language Purity & Zero Fallback
Answers strictly in the user's selected language and native writing script without defaulting back to English or Hindi.
Interactive Canvas Graphing
Safe client-side mathematical graph plotting (e.g. y = x², sin(x)) rendered dynamically without server-side compute overhead.
Visual Reference Panel
Auto-fetches educational diagrams and science images directly from the Wikipedia API based on the lesson context.
ZeroGPU Cloud Backend
Deployed on Hugging Face Spaces with dynamic GPU allocation for rapid response streaming and zero idle costs.
Clean Output (No CoT Leakage)
Internal reasoning tokens (<think>) are filtered out, leaving clean, structured answers for the student.
Supported Languages & Writing Systems
Vidya directly understands and writes in 8 major Indian writing systems across 11 languages.
Vedant Jadhav
Machine Learning Engineer • AI / LLM Researcher • Co-Founder
"Vedant Jadhav is a Machine Learning Engineer and AI/LLM researcher focused on building practical intelligent systems across language, education, and machine learning."
His work spans language models, multilingual AI, domain-specific LLMs, LLM evaluation, RAG, agentic AI, and machine learning systems.
VIDYA
"An educational AI designed to make AI-assisted learning more accessible to Indian students through multilingual education, mathematical reasoning, science, and NCERT-oriented learning."
Research
Pedagogical LLM alignment & benchmarks
Language Models
1.7B Parameter fine-tuning & SLMs
Multilingual Intelligence
11 Indian languages & zero-fallback
Educational AI
LaTeX math, canvas plots & NCERT core
VIDYA
Production AI Ecosystem for Students
Ready to Learn Deeper?
Enter the minimalist Vidya AI Playground or take a 60-second mental reset in Focus Lab.