Gayatri Chemistry Tutor
NCERT-Aligned Adaptive AI Tutoring with Deterministic Evaluators & Fine-Tuned Models
Architecture & Core Mechanics
Unlike generic LLM wrappers that hallucinate chemical equations or give shallow answers, Gayatri Chemistry Tutor pairs a neural language model with a deterministic adaptive learning engine. It features persistent student mastery tracking, structured answer evaluation for stoichiometry and thermodynamics, concept-aware RAG, and an append-only evidence ledger.
Engineering Highlights & Capabilities
-
✓
17-Stage Fine-Tuning Curriculum: Trained on Colab T4 with Qwen2.5-1.5B-Instruct using LoRA/PEFT across thermodynamics and inorganic chemistry.
-
✓
Deterministic Answer Evaluators: Custom symbolic checks for numerical tolerances, stoichiometry, formula normalization, and MCQ without naive string matching.
-
✓
Root Misconception Diagnosis: Actively diagnoses conceptual flaws (e.g. standard enthalpy of formation, logic inversion) and directs remediation.
-
✓
Portable Windows Desktop Package: Single-folder portable executable with embedded Python runtime and GGUF inference backend.
Verification & Test Telemetry
● 305 passing domain tests · 17 fine-tuning stages · Zero internet needed
Technology Blueprint
Repository Access
Access source code, architectural test harnesses, and documentation under the Gayatri-Education organization.
Inspect on GitHub →Explore Other Systems
Interested in deploying or contributing?
Our products are deployed across universities and enterprise pilots. Get in touch with our engineering team.