An Intelligent Content Intelligence Platform powered by RAG Architecture.
Vye is an AI-powered assistant designed to transform the way we consume video content. Built on a Retrieval-Augmented Generation (RAG) framework, Vye allows users to "interact" with a vast knowledge base extracted from YouTube channels, providing deep insights, summaries, and strategic analysis with zero hallucinations.
- Channel Brain Mode: Queries the specific knowledge base of a channel using vector similarity search.
- Deep Video Intelligence: Analyzes speaker tone, sarcasm, and executive-level business strategies.
- Competitor Arena: Cross-references data and perspectives between different creators or topics.
- Source Attribution: Every response includes "Sources & Evidence" to ensure factual accuracy based on transcripts.
- Responsive Interface: Custom CSS-engineered UI that adapts perfectly to mobile and desktop screens.
Vye processes long-form video transcripts into semantic chunks and stores them in a high-dimensional vector space.
- LLM: Google Gemini 2.5 Flash.
- Vector Database: Supabase with
pgvector. - Embedding Model: Gemini Embedding-001 (3,072 dimensions).
- Knowledge Depth: 3000+ Semantic chunks currently indexed.
- Framework: Streamlit for a fast, responsive web interface.
- Rate-Limit Management: Implemented customized "cooling down" periods and batch processing to optimize Google AI Studio's Free Tier limits.
- Data Pipelines: Automated transcript extraction and cleaning using
yt-dlpbefore vectorization. - UI/UX Engineering: Advanced CSS injection to handle Streamlit’s markdown containers for a modern chat experience.
See Vye in action: Witness the RAG architecture processing real-time queries.
(Click the image or the button above to play the full technical demonstration)