.png&w=256&q=75)
🤓 Latest Submissions
.png&w=828&q=75)
ARC_Studio
A.R.C Studio is a content pre-testing platform that predicts how an audience will respond to marketing content before it ships. It accepts video, audio, or text and routes each through three integrated AI systems running on a single AMD MI300X. The neural layer uses Meta FAIR's TRIBE v2, a trimodal brain-encoding foundation model. It predicts activation across seven brain regions in response to the stimulus — visual cortex, auditory cortex, language regions, limbic system, and others — producing a per-window neural timeline that surfaces exactly which moments in a video drive engagement, which fall flat, and where attention peaks. The social layer uses MiroFish, an OASIS-based multi-agent simulation engine. It builds a knowledge graph from the content, generates personas grounded in real-world entities, and runs a populated social environment in which agents read, react, share, and counter-narrate the content. The result is a behavioral signal: shares, drift, counter-narratives, sentiment trajectory. An orchestration layer powered by Qwen3.5-27B reads both signals and writes the next variant of the content, optimizing for the user's stated objective. The loop runs until thresholds clear. Built end-to-end on AMD silicon. ROCm 7.2, vLLM 0.17.1, Qwen3.5-9B for variant generation and 27B for analysis, TRIBE v2 on ROCm PyTorch. No NVIDIA in the inference path.
10 May 2026
