https://peer.madiator.cloud/videos/watch/e6057c27-e5f4-4d34-b3e6-b59eb9919af6 In this community chat, Runpod DevRel lead Jessica Garson Beauchemin sits down with staff software engineer Dean Quiñanola, one of the engineers behind Flash, to talk about how it came to be, how it works under the hood, and where it's headed next. We cover why the team replaced the build-push-redeploy Docker loop with a "portal" that ships your code straight to a serverless endpoint, the hard lessons learned from Python and NumPy version mismatches, how Dean uses Claude Code and Flash to spin up (and tear down) entire test environments for under a dollar, and what's coming as Flash merges into the Runpod Python SDK 2.0. Then Dean goes live: configuring an endpoint, hot-reloading code with flash dev, generating text and images on an RTX 4090, and having Claude Code write a brand-new Flash app from a single prompt. CHAPTERS 0:00 Welcome & introductions 1:23 Meet Dean and the origin of Flash (originally "Tetra") 2:57 Docker vs. Flash: the "portal" analogy 3:35 What is Flash? Ephemeral serverless for your code 4:36 Why Python 3.12 (NumPy + PyTorch compatibility) 5:23 Beyond Python & lessons from cloud gaming 6:56 Where Flash fits (vs. public endpoints, vLLM, ComfyUI) 8:26 Flash vs. managing infrastructure yourself 9:11 Live serverless vs. deployed apps 10:09 Why we built Flash 12:28 Why Python decorators? 13:33 Early workloads & the Flash Examples repo 15:58 Hardest bug: Python/NumPy binary mismatches 17:48 How automated testing saved the project 18:53 Building Flash with AI coding tools 21:01 Dean's Claude Code + Flash test bed 25:07 Coming soon: Flash merges into Runpod Python 2.0 26:36 Long-term vision: beyond Python 28:22 Q&A: More on the Runpod Python merge 31:35 How Flash works under the hood 32:15 Q&A: Global volumes support 33:15 Choosing GPUs in Flash 34:54 Q&A: Fastest way to load large model weights 40:59 Live demo: Configuring an endpoint 43:28 Running flash dev 44:33 Tip: Try the Flash Examples repo 45:45 Hot-reloading code & adding dependencies 49:03 Q&A: Raising your max worker limit 50:35 Demo: Text generation result 52:40 Demo: Image generation 55:03 Building a Flash app with Claude Code + Runpod MCP 56:41 Benchmark: Loading Hugging Face models on Runpod 58:20 Q&A: Do Flash and Serverless share worker limits? 1:00:26 Wrap-up & how to get involved RESOURCES 📚 Flash docs: https://docs.runpod.io/flash/overview 🚀 Flash quickstart: https://docs.runpod.io/flash/quickstart 🔄 Migrating from Flash to the Runpod SDK: https://docs.runpod.io/sdk/migrate-from-flash 💻 Flash Examples: https://github.com/runpod/flash-examples 📊 Dean's model cache benchmark: https://github.com/deanq/flash-model-cache-bench 📖 All Runpod docs: https://docs.runpod.io JOIN THE COMMUNITY 💬 Discord: https://discord.gg/runpod 🗨️ Reddit: https://www.reddit.com/r/RunPod/ This is a monthly series. Have a topic idea, or building something cool you'd like to show off? Let us know on Discord. We'd love to have you on a future episode. SPEAKERS Jessica Garson Beauchemin, DevRel Lead, Runpod Dean Quiñanola, Staff Software Engineer, Runpod Moderated by Brendan McKeag #Runpod #Python #Serverless Wed, 11 Nov 2026 17:00:34 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://peer.madiator.cloud https://peer.madiator.cloud/lazy-static/avatars/e2927aea-9e0a-454c-9327-68ec87b4da11.png https://peer.madiator.cloud/videos/watch/e6057c27-e5f4-4d34-b3e6-b59eb9919af6 All rights reserved, unless otherwise specified in the terms specified at https://peer.madiator.cloud/about and potential licenses granted by each content's rightholder.