Stories from the frontier of
live commerce AI

Two years of learnings, three internal rewrites, and one very stubborn founding team — today we ship the platform we always wanted to build.

The story behind our Text-to-Video model — why we chose Stable Video Diffusion, what we changed, and what still doesn't work.

Folders, full-text search, token dashboards, word-by-word streaming, syntax-highlighted code blocks, and a completely redesigned chat UI — RunAshChat v2 is a ground-up rebuild of everything creators asked for.

How Priya Sharma Hit 10K Concurrent Viewers on Her First AI-Hosted Fashion Drop

A Mumbai-based seller, a Diwali collection, and an AI host that handled 4,200 product questions so she didn't have to.

How We Serve 1,000 Concurrent AI Inference Requests with Sub-500ms P99 Latency

Our inference stack evolution — from a single FastAPI endpoint to a multi-region streaming pipeline processing 2M events per day.

The ₹1.05 Crore Grant That Almost Didn't Happen

How we almost missed our seed grant deadline, what the money funded, and why we'd do it differently today.

Multi-modal Emotion Recognition: Combining Audio, Vision, and Text at 30fps

Our approach to real-time emotion classification for live streams — and why we decided not to ship the first version.