PFP Labs logoPFP Labs
A hand-drawn workbench where a film strip flows through six stations: magnifier, annotation pen, vector constellation, robot arm, cloud pipeline, and wireframe screen.
TwelveLabs

Hands-on lab · Six-part series

Building with TwelveLabs: From ChatGPT to Bedrock

An AI-first learning journey: start with Jockey in ChatGPT — no code — then turn the brief into your own repository across TwelveLabs and Amazon Bedrock, one reviewed chapter at a time.

Start with part one

6 chapters92 min reading

My companion repository is the reference implementation for this series: the scripts, fixtures, and notes each chapter builds. Create your own repository to follow along.

The lab path

From a conversation to your own repo

You record the footage, direct the AI, and review every change yourself. Each chapter ends with a checkpoint commit, an exercise, and honest cleanup — costs are reviewed before the first paid call, never after.

  1. Part 01

    Build with AI First: Jockey in ChatGPT

    Chapter one of the AI-first video AI lab: a complete no-code win with the official Jockey plugin in ChatGPT — stores, uploads, corpus questions, highlight review — ending in a written brief for your AI coding partner.

    Read chapter 10 min read
  2. Part 02

    Turn the Conversation into Your Own Repo

    Build a TwelveLabs video AI repository step by step: first commit, local clips, an editable media manifest, and resume-safe managed Marengo search.

    Read chapter 25 min read
  3. Part 03

    Explain What Happened: Pegasus 1.6 Summaries and Segments

    Chapter three of the AI-first video AI lab: direct Pegasus 1.6 analysis — streamed summaries with finish-reason checks, async segmentation with explicit definitions and validated output, plus a fully implemented still-image exercise.

    Read chapter 12 min read
  4. Part 04

    Own the Retrieval and the Agent

    Chapter four of the AI-first video AI lab: direct Marengo 3.5 embeddings with dimension-safe resume state, local cosine retrieval, an implemented composed image+text query, and the Jockey API with knowledge stores, Responses, and session memory.

    Read chapter 16 min read
  5. Part 05

    Repeat the Experiment on Amazon Bedrock

    Chapter five of the AI-first video AI lab: run Pegasus 1.5 and Marengo 3.0 through Amazon Bedrock with correct response validation, complete S3 output discovery and parsing, minimum permissions, and honest cleanup.

    Read chapter 15 min read
  6. Part 06

    Make a Useful Video Workbench

    The final chapter of the AI-first video AI lab: a complete local workbench (Vite + React + TypeScript) that runs on synthetic fixtures with zero API calls, a strict privacy rule for real results, and a product chooser you can defend.

    Read chapter 14 min read