E-Mind: Translating Slides into Interactive Lecture Videos
Overview
E-Mind is an AI-powered education platform designed to bridge the gap between static slide decks and dynamic, engaging learning experiences. By orchestrating LLM agents and LlamaIndex pipelines, E-Mind automatically converts presentation slides into fully narrated, interactive lecture videos — reducing content creation time while significantly improving learner engagement.
Problem
Educators and content creators spend enormous amounts of time manually converting slides into video lectures. Static slides also fail to engage modern learners who expect interactive, multimedia-rich content. E-Mind solves both problems with an end-to-end AI pipeline.
My Role
- Architected the full LlamaIndex pipeline for slide ingestion, content understanding, and structured narration generation
- Designed and implemented custom LLM prompting strategies that increased script generation efficiency by 75%
- Fine-tuned Hugging Face language models on a custom Vietnamese educational dataset for localized narration quality
- Deployed the AI model on AWS with a scalable FastAPI backend to support concurrent video generation requests
- Delivered a solution that improved educational engagement by 50% by replacing passive slide decks with interactive video content
Tech Stack
| Category | Tools |
|---|---|
| LLM Orchestration | LlamaIndex, LLM Agents, Custom Prompting |
| Model Training | Hugging Face Transformers, Fine-tuning |
| Backend | FastAPI, Python |
| Cloud & Deployment | AWS (EC2, S3), Scalable API Infrastructure |
| Data | Custom Vietnamese Educational Dataset |
Key Results
- 🚀 75% increase in script generation efficiency via LLM-powered narration
- 📈 50% improvement in educational engagement compared to static slides
- 🌍 Multilingual support with fine-tuned models on Vietnamese educational content
- ☁️ Production-deployed on AWS with a scalable, API-driven architecture