Trinita Roy
Research Assistant, PhD Student
Room:
Office Hours: by Appointment
E-Mail: trinita[dot]roy[at]tum[dot]de
Bio
I am a PhD researcher, currently working on Personalized AI Coaching Agents in Immersive Learning Environments. My research focuses on developing immersive, LLM-driven coaching platforms that integrate multimodal interaction with enterprise data to enable context-aware, emotionally responsive, and personalized training in corporate settings.
Some of a previous works include addressing a critical and ignored trust gap in modern AI - temporal coupling, by designing a real-time privacy middleware that uses acoustic and textual triggers to let users dynamically control system behavior: pausing, masking, or deleting sensitive content mid-stream, enforced through a verifiable state machine that makes every privacy decision auditable. I also introduce a novel method of having a new family of ASR models, repurposing adversarial audio attacks, typically studied as threats to be defended against, reconceiving as a precision control mechanism, by having triggers to activate privacy-preserving behavior directly inside the ASR model itself. When triggered, the ASR variants perform surgical phrase-level masking, of sensitive spans, while leaving the rest of the transcript intact and useful.
Overall, my work is driven by a single conviction: that the most powerful AI is not the one that does the most, but the one you can actually trust.
E-Mail: trinita.roy@tum.de
LinkedIn: https://www.linkedin.com/in/trinita-roy/
GitHub: https://github.com/Trinita21
Google Scholar: https://scholar.google.com/citations?user=MXdYymEAAAAJ&hl=en
Career Development
2016 - 2022: Bachelor of Technology (B.Tech) - Engineering, WBUT, India
2020 - 2022: Master of Technology (M.Tech) - Computational Biology, IIIT Delhi, India
2022 - 2024: AI/ML Research Scientist, SciSpace, Bengaluru, India
2024 - 2025: Werkstudent Gen AI Developer, Fraunhofer Gesellschaft (IPA), Stuttgart, Germany
2024 - 2026: Master of Science (MSc) - Computational Linguistics, University of Stuttgart, Germany
2025 - 2026: Werkstudent Digital Transformation, Robert Bosch GmbH, Abstatt, Germany
2026 - now: PhD Researcher, Technical University of Munich, Germany
Research Interests
Immersive Learning, Human-AI Interaction
Gen AI, Large Language Models,
Agentic and Multimodal AI Systems
Privacy-Preserving Speech and Language Models
Responsible AI
AI Governance
Publications
ICASSP 2026 - Sensitive data Protection for Privacy aware ASR with Acoustic Triggering
AAAI 2024 - SciSpace Copilot: Empowering Researchers Through Intelligent Reading Assistance
ECIR 2024 - Harnessing AI for Effortless Scientific Discovery
ACL Anthology 2022 - Multi-document Summarization of Scientific Articles