Aditya Gautam

Aditya Gautam is an AI practitioner and leader specializing in multimodal large language models, multi-agent systems, and scalable recommendation architectures. His AI Engineer World’s Fair 2026 biography identified him as a Machine Learning Lead at Meta, where he led generative AI initiatives for Reels in user interest exploration and policy understanding. His work included architecting and training multimodal models, developing agentic solutions for adversarial video challenges, and spanning end-to-end pre- and post-training workflows and engineering pipeline optimization for large-scale production deployment. Before Meta, he spent more than three years at Google building large-scale computer vision and content understanding systems.

In his conference presentation, Gautam describes detecting unexpected content changes within short-form videos and identifying copied or transformed videos to support source attribution. He presents a three-agent architecture: a perceiver segments videos around temporal changes and extracts clip embeddings, tags, on-screen text, descriptions, and timestamps; a retriever uses inverted, vector, and graph indexes to find related clips and authors; and a reviewer combines these signals with live user reports, reactions, comments, and sentiment. He explains how specialized small vision-language models are trained on in-house visual data and tuned for structured outputs, then improved through production sampling, an in-house model judge, human review, and direct preference optimization. His approach combines distillation and quantization with evaluation of retrieval quality, latency, reasoning budgets, token costs, and judge drift. Similar-frame compression, cached decisions for viral content, and metadata-based candidate pruning reduce processing demands.

His conference biography also notes that his work has been featured by Nasdaq and Marktechpost, that he frequently speaks at events including the Databricks Data + AI Summit, Silicon Slopes, and MLOps Summit, and that he serves as a peer reviewer for NeurIPS, ICML, and AAAI. His focus connects frontier research with practical production engineering.