Soumya Sanyal

AI Researcher

Santa Clara, California, United States9 yrs 11 mos experience

Highly Stable

Key Highlights

Expert in AI safety solutions for large language models.
Strong background in risk management and equities.
Ph.D. focused on robust and explainable AI systems.

Stackforce AI infers this person is a Machine Learning and AI Safety expert in the SaaS industry.

Contact

sanyal.soumya8@gmail.com LinkedIn

Skills

Core Skills

Machine LearningSafety SolutionsExplainabilityGraph Neural NetworksRisk ManagementData Processing

Other Skills

AlgorithmsAtmel AVRCCombinatorial SearchData AnalysisData AnalyticsEquitiesEquity DerivativesExplanation-based LearningFeature SelectionFinanceHTMLHadoopInvestment BankingJava

About

I'm an Applied Scientist at Amazon Web Services (AWS), focused on building safety solutions for large language models within the Amazon Bedrock Guardrails team. My work involves tackling critical challenges like sensitive information redaction and ensuring the safety of agentic AI tools. This role is a natural extension of my Ph.D. research at the University of Southern California, where I was advised by Dr. Xiang Ren. My doctoral work was focussed on making AI systems more robust, explainable, and safe through explanation-based learning and enhanced reasoning. Throughout my Ph.D., I've had the privilege of interning at premier research labs, including Microsoft Research, where I worked on pre-training LMs for logical reasoning, and multiple terms at AWS, where I developed systems for hallucination detection and synthetic data generation. Before my Ph.D., I built a foundation in practical, high-stakes systems as a Senior Analyst at Goldman Sachs, developing risk infrastructure for the equities desk. I'm passionate about bridging the gap between cutting-edge research and real-world product safety.

Experience

9 yrs 11 mos

Total Experience

2 yrs 11 mos

Average Tenure

1 yr

Current Experience

Amazon web services (aws)

Applied Scientist

Jun 2025 – Present · 1 yr · Santa Clara, California, United States · On-site

Working on Amazon Bedrock Guardrails team on sensitive information redaction and agentic tool
safety solutions.

Machine LearningSafety SolutionsSensitive Information Redaction

University of southern california

Graduate Research Fellow

Aug 2020 – Jun 2025 · 4 yrs 10 mos · Los Angeles, California, United States · On-site

I worked on analyzing and improving various reasoning challenges in LLMs, including commonsense, deductive, and logical reasoning. I also worked in explainability and explanation-based learning algorithms.

Reasoning ChallengesExplainabilityExplanation-based LearningMachine Learning

Indian institute of science (iisc)

Research Assistant

Aug 2018 – Jul 2020 · 1 yr 11 mos · Bangalore

My main research interests lie in the domain of graph neural networks and its various applications. I'm also generally interested about NLP.

Graph Neural NetworksNatural Language Processing

Goldman sachs

Senior Analyst

Jun 2016 – Aug 2018 · 2 yrs 2 mos · Bengaluru Area, India

I work in the Equities Risk Management division in charge of managing the equities wide risk and PnL metrics. Currently, I'm working on risk engine optimizations to enhance risk metrics for structured products.

Risk ManagementEquitiesPnL Metrics

Wipro technologies

Summer Intern

May 2015 – Jul 2015 · 2 mos · Kolkata

Worked on a stand alone project to explore the use of Hadoop MapReduce infrastructure in the field of combinatorial search.

HadoopMapReduceCombinatorial SearchData Processing

Indian institute of technology, kanpur

Project Assistant

May 2014 – Jul 2014 · 2 mos · Kanpur

Developed a novel way of feature selection based on feature weights for "Machine fault detection in compressors" working under Dr. Nischal Kumar Verma. This was part of a project funded by Boeing.

Feature SelectionMachine Fault DetectionMachine Learning