Software Engineer at Galleri5 (now part of Collective Artists Network). Joined when the tech team was around 10 people, and one of four engineers building our generative AI stack — so I've had the chance to work across backend architecture, ML research, DevOps, and internal tooling.
Core work revolves around Generative AI, ML and backend systems — everything from classical techniques (NLP, clustering, vector search powering trend-based product discovery for Myntra and Ajio) to generative AI for video and image creation.
As one of the core engineers on the team, I built the backend and AI pipelines powering actual production content — including Mahabharat - Ek Dharmayudh streaming on JioHotstar. Most recently: worked with the team on a hybrid AI + 3D pipeline with multi-model orchestration designed for cinema-grade output, powering the AI-generated teaser for Chiranjeevi Hanuman: The Eternal (8M+ on YouTube / 12M+ on Instagram, premiered on IMAX).
A big part of the work involves designing autonomous AI agents that decompose creative prompts into multi-step workflows with dependency resolution, orchestrating 78 models across 7 providers (Replicate, FAL, ElevenLabs, BytePlus, Google, RunPod, GPU cluster) — plus building the internal tooling that makes these workflows usable by non-technical teams.
Infrastructure work includes building distributed systems like the Inference Gateway with WRR-based fair scheduling and 4-level concurrency control, GPU Dispatcher managing an 8 H100 GPU server fleet with lease-based locking and crash recovery via WAL. Managing deployments and migrations across GCP, Azure, and AWS with on-call rotation and monitoring.
Currently co-developing AI asset platform with Microsoft — designing multi-layer architecture for versioned storage, semantic retrieval, automated quality evaluation, and preference-aware ranking with real-time review workflows.
On the R&D side: building ComfyUI workflows, training LoRAs for character consistency and style transfer (e.g., Flux LoRAs for cloth styling in AI cataloguing), fine-tuning models, and developing new generation pipelines — from research to production.
