Headline

AI Privacy Research Engineer@Privasapien | M.Tech AI @ IISc Bengaluru | Building Safe & Reliable LLM Applications

About

I am an AI practitioner and applied researcher with over 8 years of experience spanning reliability engineering, industrial analytics, probabilistic machine learning, and scalable AI systems. My career began at Indian Oil Corporation Limited (IOCL), where I spent nearly a decade working in reliability-critical, high-availability industrial environments. Alongside leading electrical reliability operations, I architected and deployed multiple data-driven systems for predictive maintenance, anomaly detection, and operational decision intelligence. My work included probabilistic failure modeling under extreme class imbalance, calibrated risk scoring for early fault detection, and unsupervised text mining pipelines for large-scale maintenance logs—transforming raw operational data into structured, actionable intelligence Driven by a long-standing interest in machine learning, I chose to pursue an M.Tech in Artificial Intelligence at the Indian Institute of Science (IISc), Bengaluru to deepen my theoretical and research foundations. At IISc, my work spans efficient and responsible AI systems, with research contributions in optimization theory (published in IEEE Transactions on Artificial Intelligence ) and robustness in graph-structured metric spaces I also worked as a Research Intern at PrivaSapien, where I designed and productionized a gradient-based analysis framework for LLM safety and jailbreak detection, leveraging parameter-efficient strategies to reduce compute overhead while preserving robustness. In parallel, I built privacy-aware, RAG-based AI agents for real-world DPIA compliance workflows My current technical interests include: Scalable machine learning and deep learning systems Recommendation systems (collaborative filtering, graph-based models, representation learning) LLM finetuning and system building Graph neural networks Computer vision, including novel view synthesis and 3D Gaussian Splatting Responsible, privacy-aware, and deployment-aligned AI I am particularly interested in roles and collaborations that sit at the intersection of theory, systems, and real-world impact—where first-principles understanding meets production-grade engineering.

Location

Bengaluru

AI stack

claude

hugging face

ollama

huggingface

Qwen

vLLM

Work experience

AI Privacy Research Engineer at PrivaSapien, Jul 2026 - Present · 3 mos, Bengaluru, Karnataka, India, - Contributing to Turing, an AI red teaming platform, by developing a context-aware red teaming feature that analyzes business use cases, generates tailored red-teaming plans, and automates end-to-end security evaluations. - Building a domain-agnostic dynamic question generation framework that creates context-specific adversarial test cases using strategies such as Crescendo and PAIR to evaluate the safety, robustness, and reliability of LLM applications.

Career transition at Career Break, Jun 2024 - Present · 2 yrs 4 mos, Bengaluru, Karnataka, India, My early professional training at IOCL was rooted in core engineering and large-scale industrial systems. Over the last 4-5 years of my tenure, my work increasingly shifted toward data-driven problem solving, where I independently initiated and led analytics and machine learning projects to improve reliability, efficiency, and predictive maintenance across operations. Through sustained work on data-driven decision-making and predictive analytics in industrial systems, I developed a strong inclination toward data science and applied machine learning. This combination of hands-on industrial ML experience and growing interest in mathematical rigor motivated my transition to formal training in Artificial Intelligence, leading me to pursue an M.Tech in AI at IISc Bangalore, where I now focus on bridging theory, research, and real-world AI systems.

Privacy Research Intern at PrivaSapien, May 2025 - Jul 2025 · 3 mos, Bengaluru, Karnataka, India, - Designed and implemented an LLM jailbreak defense framework to detect and mitigate prompt-based adversarial attacks. - Built an AI-driven agents to assist Data Protection Officers (DPOs) in conducting Data Protection Impact Assessments (DPIAs) under the Indian DPDP Act.

Maintenance Manager ( reliability ) at Indian Oil Corp Limited, Feb 2023 - Jun 2024 · 1 y 4 mo, On-site, Developed a predictive maintenance model to identify motors at high risk of failure from historical maintenance and operating data, where failure cases were extremely scarce. The approach used learn the distribution of minority samples and generate synthetic data. This improved failure detection performance (F1 from 0.69 to 0.75). Early faults were identified by applying probability-based thresholding on model outputs, enabling proactive maintenance planning and reduced unplanned downtime.

Assistant Engineer at Indian Oil Corp Limited, Apr 2019 - Apr 2023 · 4 y, On-site, Developed a text clustering system to analyze high-volume SAP PM maintenance notifications and uncover repetitive maintenance patterns hidden in free-text descriptions. Using domain-aware preprocessing and TF-IDF–based feature extraction with K-Means clustering, the system reduced manual log analysis and provided actionable insights to support root-cause analysis, inventory optimization, and targeted mitigation of recurring issues.

Engineer at Indian Oil Corp Limited, Jul 2015 - Apr 2019 · 3 yrs 10 mos, Haldia, West Bengal, India

Education

Master of Technology - MTech, Artificial Intelligence, Indian Institute of Science (IISc)

Bachelor of Engineering - BE, Electrical Engineering, Indian Institute of Engineering Science and Technology (IIEST), Shibpur

Public links

github: https://github.com/arnabroy734