Huawei Innovation Competition
Developed and pitched an innovative AI-driven project using Huawei's advanced tech stack including MindSpore, CANN, and ModelArts. Achieved National 2nd Place.
Chokhmah AI: Personalized Learning Intelligence (PLI) Platform
Chokhmah AI is an adaptive study companion designed to solve the "Cold Start" problem in education. Unlike traditional tools that require weeks to learn a student's habits, it uses Retrieval-Augmented Imitation Learning to instantly "imitate" the expert logic found in uploaded course materials, providing immediate, grounded tutoring.
Hybrid RAG Architecture & Huawei Integration
The core of the platform is a Hybrid RAG (Retrieval-Augmented Generation) pipeline:
- Local Tensor Processing: Leverages Huawei MindSpore for efficient text chunking and retrieval logic on Ascend hardware.
- Generative Inference: Integrates Google Gemini 2.5 Flash for high-speed natural language generation.
- Backend Infrastructure: Orchestrated via Python Flask and SQLite for robust session management.
Key Features
- Intelligent Ingestion: Seamless drag-and-drop analysis for PDF, Word (.docx), and PowerPoint (.pptx) documents.
- Heuristic Adaptive Quizzing: Tracks performance per topic, automatically detects knowledge gaps, and recommends specific study actions.
- Targeted Summaries: Generates high-level course overviews or highly specific, localized summaries.
- Accessibility First: Hands-free studying powered by built-in Text-to-Speech and Voice Input.
Looking Ahead
Phase 2 development focuses on replacing heuristic rules with a Deep Reinforcement Learning Agent trained via MindSpore RL, integrating MindSpore OCR for handwritten notes, and developing dynamic knowledge graphs for visual concept mapping.