Sep 2026–present
Current
Current
Cruise
Senior Machine Learning EngineerPerception models for autonomous driving using foundation models.
Devang Kulshreshtha
Senior Machine Learning Engineer at Cruise
PhD at NYU | Prev: AWS, Mila/McGill
At Cruise, I am working on perception models for autonomous driving using foundation models. Previously, I was an Applied Scientist at AWS, where I led worked on agent governance in AgentCore and AI safety in Bedrock Guardrails.
Oct 2023–Sep 2026
Amazon Web Services
Applied Scientist II · AgentCore & BedrockAgent governance in AgentCore and AI safety in Bedrock Guardrails, including prompt-attack defense, harmful-content detection, and automated red-teaming. Introduced self-jailbreaking and mentored scientists and research interns.
Jul 2022–Sep 2023
Amazon Web Services
Applied Scientist I · Bedrock & LexASR personalization and contextual biasing.
2020–2022
Korbit.AI
Research EngineerResearch and deployment of question generation and adaptive feedback models for intelligent tutoring.
Earlier experience · Amazon, Alexa & INRIA
May–Sep 2021
Amazon Alexa
Applied Science InternImproved long-tail ASR accuracy for European locales through language-model rescoring.
2018–2020
Amazon Optimus
Machine Learning EngineerBuilt production anomaly detection for large-scale invoice monitoring.
Summer 2018
INRIA · University of Rennes
Research InternStudied subgraph mining for interpreting deep network failures and detecting incorrect predictions.
04
Education
New York University
PhD in Machine Learning
McGill University & Mila
MSc in Computer Science
IIT (BHU) Varanasi
B.Tech in Computer Science
All publications
Forthcoming
Multimodal View Training: Learning More from the Same Information
Manuscript and research details to follow.
Coming soon2026COLM 2026
Break Me If You Can: Self-Jailbreaking of Aligned LLMs via Lexical Insertion Prompting
An aligned language model can guide its own compromise: studying self-jailbreaking through lexical insertion and multi-turn search.
2026EACL 2026 (Industry)
The Subtle Art of Defection: Understanding Uncooperative Behaviors in LLM based Multi-Agent Systems
A game-theoretic framework for understanding how uncooperative behaviors destabilize collaborative agent systems.
2026EMNLP REALM Workshop 2026
STAC: When Innocent Tools Form Dangerous Chains for LLM Agents
Examining agent safety across sequences of tool calls whose cumulative effects can be harmful.
2026AACL 2026
Relational Priors as Convergence Pressure in LLM-Based Multi-Agent Systems
Explicit social relationships can encourage agent agreement without reliably improving correctness.
2024INTERSPEECH 2024
Sequential Editing for Lifelong Training of Speech Recognition Models
Sequential model editing for adapting speech recognition systems over time.
05
Patents
06
Service & recognition
Reviewing
COLM 2026 · ACL 2026 · AAAI 2026
INTERSPEECH 2024, 2025
ACL Rolling Review 2022, 2024
Teaching
IVADO Workshop, McGill / Mila
Introduction to AI, IIT Varanasi
Introduction to Programming in C, IIT Varanasi
Talks
INTERSPEECH 2023 · IJCAI-PAIS 2022
EMNLP 2021 · COLING 2018
IJCAI Workshop 2017
Recognition
MITACS Accelerate Research Scholarship
Kharusi International Science Fellowship
Shared-task winner, WASSA @ EMNLP 2017
Get in touch.
I’m always interested in conversations and collaborations around foundation models, reliable agents, and perception.
devangkulshreshtha12@gmail.com