Divyam Sengar

Discrete & Continuous Optimization

• Selected as the sole TA for a class of 100+ students, responsible for guiding students through complex optimization algorithms and mathematical formulations.

• Architected OSCAR: Developed an open-source, serverless multi-agent AI chatbot for automating OpenSearch release management. Recognized as an "AWS AI win" and presented at the OpenSearch Conference.
• Multi-Agent Framework: Engineered a sophisticated system using AWS Bedrock with agent-to-agent (a2a) protocols, coordinating a supervisor agent and specialized workflow automation agents.
• Security & Infrastructure: Implemented dual-agent security via RBAC for safe Jenkins executions. Deployed modular IaC using AWS CDK (Lambda, DynamoDB, API Gateway) adhering to least-privilege principles.
• RAG Architecture: Built a bespoke RAG system using a custom OpenSearch VectorDB to retrieve context-specific knowledge from data streams and S3 documentation.

• Optimized AI chatbot performance by designing fine-tuning strategies for transformer-based models, enhancing code correctness and efficiency.
• Engineered training pipelines for LLMs, generating diverse coding problems across multiple languages to improve model reasoning capabilities.
• Conducted differential analysis of model architectures to identify bottlenecks and implement data-driven optimizations.

• Mentored 500+ student classes for Python Programming, Advanced Data Structures, Computer Organization, and Deep Learning.
• Produced homework, exams, and dynamic question generators, significantly improving assessment of student comprehension.

• Developed a dynamic membership platform using React.js, Node.js, Express, and MongoDB, increasing membership engagement by 50%.
• Implemented secure user authentication and efficient data management for user and event data.