What I Built Locally with Qwen3.8-27B: End-to-End AI Visualizations
I have been stress-testing what is possible with an entirely local AI stack. The video below and its animations were produced with Qwen3.8-27B BF16 running on RTX 3090 GPUs, orchestrated through my local Visual Studio Code harness.
No cloud render farm, no hosted AI runtime, and no external generation service. This is the exact direction I care about: reproducible local pipelines that can create high-quality visual explainers fast while keeping full control over cost, latency, and IP.
X Video Embed
This video and the animations within were all created with qwen3.8-27b BF16 on RTX3090s + the harness Visual Studio Code. pic.twitter.com/iKQpFw02Ar
— Michel Laclé (@micheltamanda) August 26, 2026
Why This Matters
Local-first AI video workflows are becoming practical for serious production work. With the right harness and model strategy, teams can iterate quickly on storytelling, visualizations, and animation-heavy content without depending on closed hosted systems.
This is a foundation I am actively expanding into faster content systems for technical education, product demos, and client-facing explainers.
← Back to all work