Rohan Patil

Experience

1. Sr Applied Scientist Amazon Agi Org

   Amazon

   Feb 2013 — Present · Seattle, WA, US

   As a Sr. Applied Scientist at Amazon, I have built large-scale AI systems across AGI/Personal Grounding Services (PGS), music technology, Alexa, and advertising. My work spans personalization, retrieval, and advanced ML, with a focus on measurable business impact. • Current – LLM Personalization – Leading core research and development to make LLMs personalized, enabling models to adapt outputs to user history, preferences, and multimodal context. Oversee LLM supervised fine-tuning (SFT), prompt and context optimization, and end-to-end data curation pipelines from collection to evaluation. Mentor junior scientists to deliver research prototypes and production systems. • PGS & Complex Document Understanding – Delivered the retriever for multimodal, personalized agents handling complex documents and memory grounding, achieving ~90% recall and ~70% precision, with higher precision when paired with the generator. Led science vision and planning, including prototypes for structured memories, context graphs, and entity-based retrieval. • Music Catalog Intelligence – Patented large-scale audio-only clustering and duplicate detection system using DNNs, contrastive learning, and graph algorithms. Designed to handle agglomerations of tracks and process over 300M tracks in production, reducing SLA from 72 hours to 8 hours (often under 2). Achieved 90% precision and recall in human-audited evaluations of 3K+ tracks. Built LLM-based entity linkage with chain-of-thought and ReAct, improving recall by 50% with a small precision trade-off. • Advertising & Marketing Science (2017–2022) – Led projects including regression-based ad valuation (500M+ ads/day/region,+1–5% net profit lift), econometrics for marginal elasticity (+5–17%), and DNN-based bias correction (>83% bias reduction). Developed GNN+NLP cold-start valuation (+5–10% net profit) and consulted on ad valuation launches exceeding $1B in revenue. Delivered recommendation and ranking for social ads (+$10M impact).

Skills

• Spring Webflow
• Tensorflow
• Java
• Data Analytics
• Eclipse
• Keras
• Java Enterprise Edition
• Enterprise Architecture
• Ajax Frameworks
• Scikit-Learn
• Javascript
• Solr
• Subversion
• Rest
• Ubuntu
• Mysql
• Distributed Systems
• Machine Learning
• C
• Asp.net
• Hibernate
• Software Development
• Python
• Json
• Junit
• Apache Spark
• Jquery
• Web Applications
• Hadoop
• Richfaces
• Scala
• Spring Framework
• Databases
• Git
• Shell Scripting
• Design Patterns
• Representational State Transfer (Rest)
• Ajax
• Extjs
• Algorithms