Kevin Jacob

Evaluating LLMs across various bias metrics for stereotypes and robustness. Advised by Dr. Weicheng Ma

Currently working on cutting-edge projects at the intersection of machine learning, natural language processing, and assistive technologies. Key research areas include: •Developing novel methodologies to address the challenges of learning with noisy labels and annotator disagreement in computer vision and machine learning datasets. •Proposing innovative mechanisms for simulating realistic label noise in multi-annotator scenarios, enabling more robust experimentation in diverse label distribution contexts. •Investigating the impact of multi-annotator opinions on model performance across various noise types and levels. •Creating natural language models that are resilient to noise while accurately representing diverse annotator perspectives. This work aims to enhance the reliability and inclusivity of machine learning models, particularly in scenarios with subjective or ambiguous data labeling, such as content moderation and assistive technologies.

Working on AI automation tools that help our dev team move faster and share knowledge more effectively. What I've been building: LangGraph Documentation Agent: Built a microservice that watches Jira updates and auto-generates docs from GitHub PRs using semantic search. It's processing 50+ tickets per week and has cut our documentation time by 3×. We monitor everything through LangSmith to keep quality high. Cloud Infrastructure: Deployed the whole thing on Azure Functions with Docker containers and set up CI/CD pipelines. Technical writers are giving us 95% quality scores, which feels pretty good. TypeScript CLI Fixes: Spent time debugging validation bugs in our CLI tool and added schema checks so our 40+ engineers don't hit the same config issues. Secure Knowledge Platform: Architected hosted MCPs on Azure App Services with custom OAuth so 200+ people can safely access company knowledge through LLMs. Really enjoying the challenge of making AI tools that actually solve day-to-day problems for the team rather than just being cool demos.

Served as a technical advisor and implementation specialist for an advanced machine learning curriculum, ensuring complex concepts were accessible through hands-on learning. Key Contributions: •Reviewed curriculum content and provided 100+ actionable suggestions across 9 textbook chapters • Built and validated feasible ML projects using Python and Scikit-learn for diverse student skill levels • Pioneered new hands-on projects in emerging technologies to strengthen the practical component Impact: Played a crucial role in modernizing the machine learning curriculum by introducing industry-relevant, executable projects while maintaining accessibility for 300+ students.