Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
May 18, 2026
·
San Francisco
AI Video Editing Pipeline
Learn how to build a deterministic AI video editing pipeline using LLMs for smart decisions and ffmpeg for reliable rendering, breaking down complex tasks into small, tested skills.
Overview
AI Tinkerers founder Joe Heitzeberg explains a deterministic multi-agent pipeline that automatically detects and edits talks, cleans up audio, makes transcript-aware edit decisions, and uses validated JSON plus ffmpeg for reliable rendering.
Tech stack
- Deepgram Nova-2Deepgram Nova-2 is a lightning-fast speech-to-text API that slashes word error rates by 30% compared to competitors while keeping costs at a fraction of a cent per minute.Engineered for high-throughput enterprise demands, Deepgram Nova-2 optimizes the standard Transformer architecture to deliver speech-to-text transcription with unmatched speed and accuracy. It achieves a 36% relative word error rate improvement over OpenAI Whisper (large) and processes an hour of pre-recorded audio in just 12 seconds. By combining advanced data curation with a multi-stage training methodology, Nova-2 excels at tricky formatting, proper nouns, and filler word identification (all for a highly competitive starting rate of $0.0043 per minute).
- ClaudeClaude is Anthropic's flagship family of large language models (LLMs): a high-performance, Constitutional AI system built for safety, complex reasoning, and expert-level collaboration.Claude is a next-generation AI assistant developed by Anthropic, a research firm prioritizing AI safety. The models (including Opus, Sonnet, and Haiku) leverage Constitutional AI to ensure helpful, honest, and harmless outputs, a key differentiator from competitors. Claude excels at complex enterprise tasks: processing massive context windows for in-depth data analysis, generating and reviewing code, and providing expert-level summarization for documents up to 200,000 tokens. It is deployed as a conversational chatbot and via API, offering scalable AI solutions for developers and businesses.
- RailsRuby on Rails is a full-stack web application framework designed to accelerate development through sensible defaults and a convention-over-configuration philosophy.Built on the elegant Ruby language, Rails provides everything a developer needs to build database-backed web applications from day one. It champions developer happiness and rapid shipping by eliminating boilerplate code, relying instead on established conventions (like the Model-View-Controller architecture) to handle routing, database migrations, and asset management automatically. This highly productive approach has made Rails the foundation for industry giants like GitHub, Shopify, and Airbnb, proving that a single developer can build and scale a massive product quickly.
- SidekiqSidekiq is a multithreaded background job processor for Ruby that uses Redis to handle millions of tasks with minimal memory overhead.Sidekiq keeps Ruby applications fast by moving slow, heavy tasks (like processing images, sending emails, or calling third-party APIs) out of the main request-response cycle. By leveraging Redis for fast data storage and utilizing a multithreaded architecture instead of a multi-process one, a single Sidekiq process can run dozens of concurrent jobs, saving massive amounts of RAM compared to older engines like Delayed Job. It integrates directly with Ruby on Rails, scales to thousands of processes and billions of jobs per day, and includes a built-in web dashboard to monitor queue health in real time.
- ffmpegFFmpeg is the leading, open-source multimedia framework: it decodes, encodes, transcribes, muxes, demuxes, streams, and filters almost any audio or video format.FFmpeg is a powerful, cross-platform command-line utility and library suite (LGPL/GPL) for processing multimedia content. The core `ffmpeg` tool handles universal format conversion—like transcoding an input MP4 to an output WebM—while bundled utilities like `ffprobe` provide deep stream analysis and `ffplay` offers simple playback. Its foundational libraries, including `libavcodec` (for codecs) and `libavformat` (for containers), support thousands of formats and protocols, making it indispensable for major platforms like YouTube and VLC for everything from real-time streaming to complex video filtering.
Finding related talks...
Compose Email
Sending...
Email preview
Loading recent emails...