I'm a software engineer in Bengaluru at Galleri5 (part of Collective Artists Network). I build generative AI systems and the infrastructure they run on. I joined when the tech team was about 10 people, and as one of four engineers on the AI stack I work the whole depth of it: ML research and model training at one end, distributed systems and production ops at the other. Most of that in my first 3 years.
Most of what I do is architect and build our canvas: an agentic system that turns a plain creative brief into finished images, video, and audio. Under it is a harness that drives generation end to end. You describe what you want, and it plans the work: picks the right model for each step, holds one route per turn so nothing drifts, and checks with you before it burns real compute. Give it a video brief and it goes all the way to a stitched cut. I'm as much an infra person as a dev, so I own what runs under it too: the H100 and RunPod fleet, the job dispatch and crash recovery, the deploys across GCP, Azure, and AWS.
The pipelines I built went into Mahabharat - Ek Dharmayudh on JioHotstar, and the AI teaser for Chiranjeevi Hanuman: The Eternal, which premiered on IMAX and did 8M+ views on YouTube and 12M+ on Instagram. I'm also co-developing an AI asset platform with Microsoft.
The canvas is only the last year. Before it I was hands-on with the models themselves: image and video pipelines in ComfyUI and PyTorch, CUDA where it mattered, LoRAs trained for character consistency and style transfer. That's the craft the canvas orchestrates now, and I still spend a lot of my time in it on the R&D side. Earlier than that it was classical ML, the NLP and vector search behind product discovery for Myntra and Ajio.
