# QuickVid

> QuickVid by Ayan Parkar summarized, fact-checked and answered
> questions about any YouTube video with transcripts, Gemini and RAG.
> Featured by LangChain, 174 GitHub stars.

[← AYAN.WTF](https://www.ayn.wtf/) \ PROJECTS \ QUICKVID

## Overview

Summarize, verify, and chat with any YouTube video. You paste a link and
get a summary with notes in about 10 seconds. Then you can check the
video's claims against the web, or ask it questions.

- TYPE: Web app
- STACK: Next.js, Neon, LangChain, Gemini
- YEAR: 2024
- STATUS: Shut down
- REPO: [metaloozee/quickvid](https://github.com/metaloozee/quickvid)

## How it works

I started with Whisper and GPT in 2023. That version downloaded each
video's audio and ran it through speech-to-text, which took minutes.

I rebuilt it around YouTube's own transcripts and Gemini. QuickVid
splits a transcript into 5,000-token pieces, has Gemini 1.5 Flash
summarize the pieces in parallel, then merges them into one summary with
bullet-point notes. Long videos work the same way as short ones, there
are just more pieces.

## Verify

A summary of a wrong video is still wrong. So QuickVid can fact-check
one.

Gemini writes five search queries from the summary and Tavily pulls web
results for each. The model then compares the summary with those sources
and returns a verdict, accurate or not, with an explanation of how it
got there.

## Chat

For chat, QuickVid cuts the transcript into 250-token chunks, embeds
them and stores the vectors in Postgres on Neon. Each question pulls the
closest chunks by cosine similarity and hands them to Gemini, so answers
come from what was said in the video.

## Results

LangChain featured it. 174 GitHub stars, 15 forks, 100+ active users.
Summarization went from minutes to about 10 seconds.

## Why it ended

In August 2024 the deployed app started failing on YouTube requests. I
tried YouTube cookies, then proxies, then swapped ytdl for youtubei.js.

YouTube ToS and crawler blocking made the workflow unsustainable, so I
shut it down. Still the most successful side project I've shipped.

## More projects

- [DITTO](https://www.ayn.wtf/projects/ditto): A durable AI coding
  workspace on the web. Each GitHub repo gets its own sandbox with
  parallel sessions: isolated git worktrees, resumable agent runs over
  SSE, and a real commit-to-PR workflow shared by agents and humans.
  Still in active development.
- [RANKMESH](https://www.ayn.wtf/projects/rankmesh): An autonomous SEO
  and AI-search visibility engine: agents audit, fix, write and publish
  while people review and approve. Multi-service FastAPI backend,
  Celery/Redis jobs with Claude and Codex runners, and a Next.js panel.
  This is what I ship day to day.
- [BACKPACK](https://www.ayn.wtf/projects/backpack): Self-hostable,
  open-source AI research and chat app in the spirit of NotebookLM: 6+
  LLM providers, RAG research spaces, cross-chat memory, configurable
  MCP servers, web and academic search, and a versioned artifact
  workspace.

[← BACK TO AYAN.WTF](https://www.ayn.wtf/)

---
Screen font: Px437 IBM VGA 8x16, Ultimate Oldschool PC Font Pack
(int10h.org), CC BY-SA 4.0.