Shubhanshu Chauhan

Senior Principal Scientist @Eli Lilly and Company

Bengaluru, IN

Led the technical design for a multi-agent framework to automate the synthesis of complex, domain-specific reports. The architecture is focused on solving core challenges in scalable reasoning, long-term adaptation, and human-AI collaboration.System Architecture: Designed a multi-agent reasoning system using LangGraph where specialized agents collaborate on cognitive tasks. The architecture features a Deep Planning Agent that interprets ambiguous requirements and translates them into a verifiable, machine-executable flow, improving the reliability of the generative process.Scalability and Performance: Developed a high-performance computation model using asyncio that decouples the system\'s orchestration logic from the task volume, achieving ~O(1) time complexity for internal scheduling. This design ensures overall throughput scales linearly with available API bandwidth, and under typical limits, it reduced end-to-end latency by over 80%.Agent Memory Design: Prototyped a hierarchical memory architecture designed for continuous, guided learning. The model includes a two-tiered system with agent-specific and global strategic memories. It proposes an LLM-based \"Judge\" for self-curated memory retention, with built-in checkpoints for periodic human-in-the-loop validation and alignment.Resilience and State Management: Implemented a robust hybrid checkpointing system for long-running, interactive AI workflows using AWS DynamoDB and S3. This ensures fault tolerance and allows for seamless human intervention, making the system a durable and collaborative tool.

Indian Institute of Technology, Kanpur

MSc, Physics

Indian Institute of Science Education and Research (IISER), Pune

Doctor of Philosophy - PhD, Elementary Particle Physics