Devansh Kumar

WORK HISTORY

EDUCATION

ABOUT DEVANSH KUMAR

I build agentic AI systems at IBM across two core initiatives: Project BOB, where I design RAG pipelines, evaluation frameworks, and LLM workflows for a developer-focused AI assistant for IBM I customers; and the MCP Server, where I develop secure agents, authentication workflows and agent evaluation frameworks.My journey has taken me across research, industry, and teaching. At Huawei Canada, I developed agentic AI systems for factory automation, cutting inference time by 95% and contributing to a presentation at GLOBECOM’24. During my master’s at the University of Ottawa, I partnered with Accenture R&D to fine-tune embedding models and improve RAG frameworks for text-to-SQL, boosting query accuracy by over 20%. Earlier, at Samsung R&D (PRISM), I enhanced Bengali NLU models and reduced ASR latency, directly improving customer interactions.Beyond technical work, I’ve been a teaching assistant and a team lead, r—roles that sharpened my ability to mentor, collaborate, and communicate complex ideas effectively.What drives me is the intersection of AI research and practical applications,whether that’s optimizing large language models, designing retrieval systems, or creating tools that empower others. I thrive on challenges that require both technical depth and creative problem-solving.