Headline

GenAI | Backend | Infra @galleri5

About

Portfolio: https://ausafmo.com G5 AI Studio: https://aistudio.galleri5.com AI platform engineer at Galleri5 (part of Collective Artists Network), working on G5 AI Studio, a production AI content platform for ads, microdramas, shows, and films. I build and operate agentic AI systems end to end: brief interpretation, workflow planning, specialist-agent orchestration, generation, editing, validation, and long-running execution. My work sits across product and infrastructure, from the agent runtime down to the inference layer it runs on. These systems have powered production work including AI Mahabharat on JioHotstar and Chiranjeevi Hanuman, which premiered on IMAX and crossed 8M+ YouTube views. Two systems I own: - Agentic workflow runtime for G5 AI Studio: turns a creative brief into an executable production flow, covering planning, specialist-agent orchestration, a dynamic tool registry, typed tool execution, evals and validators, streaming state, and human-in-the-loop review. - Inference Gateway: grew out of a GPU dispatcher I built, now routing 80+ models across 7 providers, including external APIs and internal H100 capacity, with WRR scheduling, failover, lease-based concurrency, retries, and WAL recovery. I've also workd on AI asset platform with Microsoft: versioned storage, semantic retrieval, automated quality evaluation, preference-aware ranking, and real-time review workflows. Tech: Python, FastAPI, PyTorch, Microsoft Agent Framework, Gemini, Claude, OpenAI, ComfyUI, MongoDB, Redis, Pinecone, Docker, GCP, Azure, AWS, RunPod.

Location

Bengaluru

AI stack

claude code

codex

cursor

notebooklm

opencode

RunPod

Work experience

Manager - Software Engineering at Galleri5 (now, part of Collective Artists Network), Jun 2024 - Present · 2 yrs 3 mos, Bengaluru, Working on G5 AI Studio, Galleri5's production AI content platform for ads, microdramas, shows, and films. Built the end-to-end agentic workflow module that turns creative briefs into executable production flows: planning, specialist-agent orchestration, dynamic tool registry, typed tool execution, evals/validators, streaming state, and human-in-the-loop review. Built the GPU Dispatcher that evolved into the Inference Gateway, now routing 80+ models across 7 providers including external APIs and internal H100 GPU capacity. Handles WRR scheduling, provider failover, Redis queues, lease-based concurrency, retry handling, WAL recovery, monitoring, and manifest-driven onboarding. Contributed to production AI content workflows including AI Mahabharat on JioHotstar and Chiranjeevi Hanuman, which premiered on IMAX and crossed 8M+ YouTube views. Co-developing an AI asset platform with Microsoft: versioned storage, semantic retrieval, automated evaluation, preference-aware ranking, and real-time review workflows. Earlier: built ML/data pipelines for Galleri5's Social Intelligence and Trends Intelligence products, including semantic product search across millions of SKUs, sentence-transformer classification, Pinecone upsertion, async ingestion, trend discovery, product curation, deduplication, clustering, and forecasting.

Machine Learning Engineer Trainee at Galleri5 (now, part of Collective Artists Network), Dec 2023 - May 2024 · 6 mos, Bengaluru, Karnataka, India

Education

Bachelor's degree, Computer Science, kiet group of institutions

Public links

leetcode: https://leetcode.com/MoAusaf

github: https://github.com/ausafmo