DotLDR

TL;DR — Smart course generation and video summarization powered by Gemini AI.

                        

STOP
WATCHING.
START
KNOWING

with

TL;DR

Demo

Architecture

       YouTube URL
            │
            ▼
    [ Transcripting ] ───▶ youtube-transcript
            │
            ▼
    [   Chunking    ] ───▶ Token-aware (Gemini-optimized)
            │
            ▼
    [      RAG      ] ───▶ Semantic search (In-memory)
            │
            ▼
       [ AI Tutor ] ───▶ Gemini 2.0 Flash

The Pipeline

1. Transcripting: We extract raw text and timing data from YouTube videos using high-reliability transcript fetching.
2. Chunking: Large transcripts are split into semantically dense chunks. Our token-aware chunker targets ~1,000 tokens per segment to maximize context for Gemini while staying within efficiency limits.
3. RAG (Retrieval-Augmented Generation): For the interactive chat tutor, we use a custom RAG pipeline. It performs semantic search over lesson chunks using vector embeddings to retrieve the most relevant context for every student question.

Core Services

• Models: Powered by Gemini 2.0 Flash and 2.5 Flash for lightning-fast reasoning and course generation.
• OTP Gateway: Secure authentication via a Cloudflare Worker gateway integrated with the Resend API for high-deliverability email verification.
• Database: MongoDB serves as our primary persistence layer for course structures, progress tracking, and chat histories.

Built with intention.