Pradyoth P

Researched, designed and developed a toolkit to conduct Gen AI Red Teaming Exercises for all AI use cases in JPMC.
Led several Red Teaming Assessments for top AI initiatives at JPMC uncovering multiple critical and high-severity vulnerabilities across prompt injection, data leakage, system prompt leakage, intent drift, misuse scenarios and more.
Set up Network Broker to facilitate connection between the AI Red Teaming App and the AI Red Teaming External Vendors Instance.
Updated the application with middlewares to establish connectivity to the AI use cases.
Designed several new ECS tasks to expand red teaming coverage with multiple tools.
Designed advanced manual red teaming methodologies for agentic AI systems to uncover behavioral vulnerabilities, privilege misuse, intent misalignment, and multi-step exploitation paths beyond automated evaluations.
Analysed and evaluated coding sandboxes from a security perspective during an internal hackathon.
Consistently seen as the top contributor on Red Team Assessments, uncovering more findings than the rest of the team combined.

Researched, Designed and Developed a toolkit to conduct Gen AI Red Teaming Exercises for all AI use cases in JPMC., Led several Red Teaming Assessments for top AI Initiatives at JPMC uncovering multiple critical and high-severity vulnerabilities across prompt injection, data leakage, system prompt leakage, intent drift, misuse scenarios and more., Set up Network Broker to facilitate connection between our AI Red Teaming App and the AI Red Teaming External Vendors Instance., Updated the application with middlewares to establish connectivity to the AI Use cases., Designed several new ECS Tasks to expand red teaming coverage with multiple tools., Designed advanced manual red teaming methodologies for agentic AI systems to uncover behavioral vulnerabilities, privilege misuse, intent misalignment, and multi-step exploitation paths beyond automated evaluations., Analysed and Evaluated Coding Sandboxes from a security perspective during an internal hackathon., Consistently seen as the top contributor on Red Team Assessments, uncovering more findings than the rest of the team combined.

Researched, designed and developed a model scanning service application to conduct procurement and periodic model scanning for all externally sourced models used in JPMC (~2500+ procured models).
Took ownership of production releases, performed UAT validations, and coordinated with Production Support to release application and infrastructure updates.
Migrated the application and infrastructure deployments from Jenkins to Spinnaker as part of app modernization.
Authored new automation scenarios, reduced end-to-end test execution by ~67% to accelerate feedback, and enforced secure-coding standards.
Implemented an automated Kafka-based messaging system to identify and raise findings for non-compliant artifacts, ensuring efficient and real-time communication.
Created a robust security report for the Model Scanning Tool and developed a Model Scoring Mechanism for a new LLM Security Inference Testing Tool.
Developed a security-aligned scoring mechanism over outputs from LLM vulnerability scanner to prioritize risks and integrate with review workflows.

Researched, Designed and Developed a model scanning service application to conduct Procurement and Periodic Model Scanning for all Externally sourced Models used in JPMC (~2500+ procured models)., Took ownership of Production Releases, performed UAT validations, and coordinated with Production Support to release application and infrastructure updates., Migrated the Application and Infrastructure deployments from Jenkins to Spinnaker as part of app modernization., Elevated SDLC quality: authored new automation scenarios, reduced end-to-end test execution by ~67% to accelerate feedback, enforced secure-coding standards., Implemented an automated Kafka-based messaging system to identify and raise findings for non-compliant artifacts, ensuring efficient and real-time communication., Created a robust Security Report for the Model Scanning Tool and developed a Model Scoring Mechanism for a new LLM Security Inference Testing Tool., Developed a security-aligned scoring mechanism over outputs from LLM vulnerability scanner to prioritize risks and integrate with review workflows.

Led the SEP Engineer Committee at JPMorgan Chase's Bengaluru Tech Centre, and worked alongside the Tech Centre Leadership to curate impactful events, inspirational talks and professional training for early career software engineers.
Facilitated development of technical and professional skills for over 1100 early career employees, enhancing their learning journey and career growth.

Achieved over 60% AWS cost savings by implementing Lightswitch policies, as well as Janitor and S3 Lifecycle policies.
Integrated external vendor APIs to extend the application's capabilities and cover more use cases.
Provisioned infrastructure using Terraform to set new policies, upgrade deprecated modules, and standardize EKS, EBS, and S3 across environments.
Troubleshooted and resolved PROD bugs to ensure more accurate outputs, and fixed major vulnerabilities.
Ensured 80% code coverage by adding unit test cases for new functionalities.

Achieved over 60% AWS Cost Savings by implementing Lightswitch policies, as well as Janitor and S3 Lifecycle policies., Integrated external vendor APIs to extend the application's capabilities and cover more use cases., Provisioned infrastructure using Terraform to set new policies, upgrade deprecated modules, and standardize EKS, EBS, and S3 across environments., Troubleshooted and resolved PROD bugs to ensure more accurate outputs, and fixed major vulnerabilities., Ensured 80% code coverage by adding unit test cases for new functionalities.