I'm an AI Researcher, currently completing my Master's degree at Northeastern University's AI program from Khoury College of Computer Sciences (GPA: 4.0/4.0). My work sits at the intersection of contributing new knowledge to AI community and bringing real-world impact, with a focus on Foundation Generative Models, Model Interpretability, Theory of Mind in AI, and production-grade machine learning systems.
Currently, I'm working with Dr. Agata Lapedriza Garcia (Advisor), and Dr. Natalie Shapira on testing VLMs' Theory of Mind capacity, while also architecting production RAG systems that serve thousands of users.
- Vision-Language Models & Theory of Mind: Developing taxonomy and experimental frameworks for evaluating model cognition
- Kolmogorov-Arnold Networks: Investigating learning and generalization capabilities
Institute for Experiential AI | AI Research Assistant & Data Scientist
- Architected production-grade RAG system with 90% latency reduction serving 10K+ users
- Achieved 40% improvement in retrieval precision using AWS Lambda, EC2, FAISS, and OpenSearch
- Deployed XGBoost classification model with 85% F1-score on highly imbalanced datasets (10:1 ratio)
Amazon | Application Engineer III
- Reduced operational ticket volume by 35% through strategic automation
- Built ETL pipelines and QuickSight dashboards enabling data-driven resource optimization
- Maintained 99.9% uptime for production data systems serving Amazon Exports Organization
- Agents of Chaos (Red-teaming study on AI agents)
- Lead Author: Natalie Shapira
My contribution: - CS2 (Non-Owner Compliance): Probed whether agents follow instructions from non-owners — showed agents complied with exposing emails and its contents of all agent interactions.
- CS3 (Disclosure of Sensitive Information): Showed Jarvis’s PII refusal could be bypassed by reframing “share” as “forward,” exposing SSNs, bank data, and medical records via a trivially different request; one of the paper’s clearest vulnerability findings.
- CS14 (Data Tampering Refused): Tested whether Jarvis could be pressured to modify source data to cover up the PII exposure; it held firm — a documented safety success, with the agent maintaining its API boundary under persistent social pressure.
Leading end-to-end MLOps pipeline for radiological scan-based disease prediction with integrated QA RAG system for medical report analysis. Implementing DVC, TFDV, MLFlow, and GCP deployment orchestrated with Apache Airflow.
Developed post-hoc answer attribution models for long-document comprehension, leveraging Vectara's HHEM model to advance trustworthy RAG and QA systems through improved source grounding.
Fine-tuned GPT-2 on emotion classification, analyzing attention mechanisms through head masking and token replacement. Developed visualization methods for transformer attention patterns.
Hands-on implementations and experiments with coding foundation models—building and training models for code understanding and generation. Currently implemented VAEs. Coding Diffusion Models for the next push.
Implemented Q-Learning with multi-agent training in custom gymnasium environment, optimizing state representation as ternary sequences for efficient training.
Fine-tuned ALBERT and DeBERTa-XS transformers achieving >96% accuracy in detecting AI-generated content on DAIGT dataset.
https://github.com/rjaditya-2702?tab=repositories
Northeastern University | Master of Science, Artificial Intelligence | GPA: 4.0/4.0
Khoury College of Computer Sciences | Sept 2023 - Dec 2025
Vellore Institute of Technology | Bachelor of Technology, ECE | GPA: 9.11/10.0
Chennai, India | July 2017 - June 2021
Teaching Assistant - CS5100: Foundations of AI (Spring '24 & Fall '24)** Supported students with conceptual questions, assignments, and project guidance. (Oooh, I delivered a full lecture on Gradient Descent. It was fun! and I absolutely would do it again!)
💡 Open to collaborations in AI research, Machine Learning, Multi-modal generative AI, and Interpretability