Arvind Srinivasan

Co-Founder

Palo Alto, California, United States4 yrs 2 mos experience

Key Highlights

Expert in multimodal foundation models and representation learning.
Pioneered large-scale retrieval and ranking innovations at Amazon.
Co-authored notable research papers in LLMs and data quality.

Stackforce AI infers this person is a highly skilled AI professional specializing in multimodal models and large-scale machine learning applications.

Contact

Skills

Core Skills

Representation LearningReinforcement LearningMultimodal Embedding ModelsKnowledge DistillationMulti-objective OptimizationMachine Learning Algorithms

Other Skills

Large-Scale RetrievalPost-TrainingSynthetic Data StrategiesQuantizationLoRABERTNatural Language ProcessingClustering AlgorithmsEnd-to-End ML PipelineData ScrapingVisualization ToolsFlaskGraph-Based ML AlgorithmsSparse-Dense Matrix MultiplicationMobile Application Development

About

I am an Applied Scientist II at Amazon, building multimodal foundation models through representation learning and reinforcement learning. I focus on developing billion-parameter scale LLMs optimized for Search and Advertising, powering large-scale applications in retrieval, ranking, and ad relevance. My work spans multi-task training, multi-objective optimization, post-training (SFT, RL), and efficient training/deployment using LoRA, quantization, and knowledge distillation. These models are widely adopted across Amazon Search and Ads systems, supporting diverse applications from query–product retrieval to ad-click prediction. I hold a Master’s degree in Machine Learning from Carnegie Mellon University, where I co-authored papers on time-series + text LLMs and data quality assessment (NeurIPS, AAAI). My work combines research-driven innovation with production-scale deployment, and I am particularly focused on the future of foundation models — multimodality, efficiency, reasoning, and alignment.

Experience

4 yrs 2 mos

Total Experience

1 yr 3 mos

Average Tenure

4 mos

Current Experience

Unity

Senior Applied Research Scientist

Jan 2026 – Present · 4 mos · Mountain View, California, United States · On-site

Amazon

2 roles

Applied Scientist II

Promoted

Oct 2024 – Jan 2026 · 1 yr 3 mos · On-site

Tech lead for representation learning with LLMs, pioneering innovations in large-scale retrieval and ranking, including:
▪ Post-training of 1B-8B LLMs with RL and SFT for query/product classification tasks, achieving Claude-level performance at 20x throughput.
▪ Synthetic data strategies for noisy-labeled datasets and data diversity, improving retrieval quality for challenging natural-language queries.
▪ Reasoning-driven agentic embeddings and generative–representation hybrid LLMs for advanced query understanding.

Representation LearningReinforcement LearningLarge-Scale RetrievalPost-TrainingSynthetic Data Strategies

Applied Scientist

Aug 2023 – Oct 2024 · 1 yr 2 mos · On-site

▪ Trained multimodal embedding models (1B–32B) supporting Search and Ads systems in product retrieval, ranking, and ad click-through prediction.
▪ Designed a multi-objective contrastive loss optimizing for relevance, clicks, and purchases, outperforming strong baselines in internal evaluations.
▪ Delivered production models through knowledge distillation and quantization, enabling low-latency deployment at scale.
▪ Scaled training with LoRA+, distributed training, and mixed precision, significantly improving efficiency of billion-parameter model training.

Multimodal Embedding ModelsMulti-Objective OptimizationKnowledge DistillationQuantizationLoRA

Carnegie mellon university robotics institute

Research Programmer

Feb 2023 – Aug 2023 · 6 mos · Pittsburgh, Pennsylvania, United States · On-site

▪ Advised by Prof. Artur Dubrawski, Auton Lab.
▪ Developed the first joint time-series + text LLM for ECG interpretation (NeurIPS 2023 Workshop Paper).
▪ Published a benchmark on label error detection (NeurIPS 2023 Paper).

Amazon

2 roles

Applied Scientist Intern

May 2022 – Aug 2022 · 3 mos · Palo Alto, California, United States

▪ Representation learning using BERT-style models for products on amazon.com.

Representation LearningBERTNatural Language Processing

SDE - Machine Learning

Jul 2020 – Jun 2021 · 11 mos · Bengaluru, Karnataka, India

▪ Developed ML algorithms for detection of robotic traffic on amazon.com.
▪ Implemented a large-scale clustering algorithm using Spark and Tensorflow. Improved incumbent production model training time by 40%.
▪ Built representations for Amazon customers from website activity using sequence modeling
techniques.
▪ Deployed an end-to-end ML pipeline for real-time classification of robot traffic.

Machine Learning AlgorithmsClustering AlgorithmsEnd-to-End ML Pipeline

Mathworks

Development Intern

Jan 2020 – May 2020 · 4 mos · Bengaluru, Karnataka, India

▪ Worked in the Engineering Development Group
▪ Developed a tool to scrape and process MATLAB 3p library metadata using Perl
▪ Built a visualization and analytics tool for MATLAB 3p library metadata using Flask

Data ScrapingVisualization ToolsFlask

Carnegie mellon university

Research Intern

Jun 2019 – Jul 2019 · 1 mo · Pittsburgh

▪ Analyzed performance bottlenecks in graph-based ML algorithms, advised by Prof. Tze Meng Low.
▪ Designed a sparse-dense matrix multiplication algorithm that outperformed the computation
speed of industry standard (Intel Math Kernel Library) by 20%.
▪ Presented “Accelerated Learning on Graphs” poster at ECE Summer Intern Research Symposium,
Carnegie Mellon University.

Graph-Based ML AlgorithmsSparse-Dense Matrix Multiplication

Indian institute of science (iisc)

Research Intern

Jun 2018 – Jul 2018 · 1 mo · Bangalore, India

▪ Worked in the department of Electronics Systems Engineering under Prof. H. S. Jamadagni
▪ Developed a mobile application to extract usage statistics from android smartphones
▪ Studied smartphone usage to analyze human behavior

Mobile Application DevelopmentUsage Statistics Analysis

Ittiam systems pvt ltd

Intern

May 2017 – Jul 2017 · 2 mos · Bengaluru, Karnataka, India

▪ Worked in the Computer Vision and Machine Learning Division
▪ Developed an Analysis tool for Incoherent Data in Ittiam’s Retail Analytics product – IttiamInsite.
▪ Designed a Web Application using the Django Framework for system administrators.
▪ Currently deployed in over 200 retail stores around the world.

Data AnalysisWeb Application DevelopmentDjango