VidIntel

Rewatching long videos and meeting recordings to extract key decisions and action items is tedious and time-consuming. Users need a way to quickly access structured insights from video content without manually scrubbing through hours of footage.

VidIntel accepts video URLs or file uploads, transcribes audio using Groq Whisper (English) and Sarvam AI (Hinglish), then runs parallelized Mistral AI LLM extraction to generate summaries, decisions, and action items. A ChromaDB-powered RAG pipeline lets users chat directly with transcript context. Server-Sent Events stream real-time processing updates, and Appwrite-based smart caching bypasses the LLM pipeline for previously processed videos.

• Universal Input — Seamlessly handles web URLs (via yt-dlp) and direct file uploads (MP4, MP3, WAV, M4A)
• Lightning Fast Transcription — Groq Whisper API for English, Sarvam AI API for native Hinglish support
• Smart Summarization — Mistral AI LLM generates concise executive summaries
• Action Item Extraction — Automatically pulls Next Steps, key decisions, and open questions
• RAG Video Chat — Embeds transcripts via Mistral Embeddings into ChromaDB Vector DB for context-aware Q&A
• Real-Time Processing Updates — Server-Sent Events stream live progress (Downloading → Transcribing → Analyzing)
• History & Cloud Sync — Appwrite-powered history dashboard with secure cloud storage
• Smart Caching — Bypasses LLM pipeline by reloading past analyses from Appwrite for repeated videos

• Built an advanced RAG architecture embedding and searching transcript data via ChromaDB for precise video Q&A
• Engineered a real-time SSE streaming pipeline for live progress updates across multi-stage processing
• Implemented parallelized LLM extraction for audio chunking, transcription, and structured insight generation
• Designed a premium glassmorphic dark-mode UI with React 19, Vite, and custom CSS micro-animations
• Built smart caching layer with Appwrite to bypass redundant LLM processing for previously analyzed videos

• 01

  Orchestrating a multi-stage async pipeline (download → chunk → transcribe → analyze) with real-time SSE feedback

• 02

  Designing a RAG architecture that produces precise, context-aware answers from long-form transcript data

• 03

  Handling dual transcription engines (Groq Whisper for English, Sarvam AI for Hinglish) with seamless fallback

• 04

  Building a smart caching system that reliably detects and reloads prior analyses from Appwrite cloud storage