2024-09-28 –, AI/SRE
As enterprises rapidly adopt generative AI (GenAI) technologies, DevOps teams face a new set of challenges and responsibilities. Large Language Model Operations (LLMOps), a subfield of MLOps, has emerged to manage the complexities of deploying and maintaining large language models (LLMs).
LLMOps requires a collaborative effort across data science, DevOps, and IT teams, emphasizing the importance of integrating these functions to ensure the seamless operation of LLMs in enterprise settings. Attendees will gain insights into best practices for managing the unique demands of LLMs, including strategies for maintaining model accuracy, scalability, and compliance with ethical standards.
I will be covering the essential practices and workflows for developing and deploying large language models efficiently. This session will cover the following key topics:
What is LLMOps?: Understand the concept of LLMOps and its role in the lifecycle of AI models, from development to production.
Development and Production Workflow: Dive into the structured approach of LLMOps, focusing on how it streamlines the development and deployment processes for AI models.
Necessity and Advantages of LLMOps: Learn why LLMOps is crucial for organizations leveraging AI, including its benefits such as improved model management, scalability, and faster iteration cycles.
Components of LLMOps: Explore the key components that make up LLMOps such as prompt engineering, model fine-tuning, review and governance, inference management, and monitoring with human feedback.
Best Practices in LLMOps: Discover best practices that ensure effective implementation of LLMOps, iterative data handling, usage of open-source libraries for fine-tuning, continuous model monitoring and Regulatory Compliance
Binoy Thomas is an expert in DevOps and SaaS deployment architectures, with a specialized focus on Vector Database (VectorDB) . Being part of Data and AI department, having extensive experience in validating RAG(Retrieval augmented generation ) with multiple embedding models and VectorDB indexing. With a track record of contributing to innovation and patenting, filed multiple USPTO applications(waiting on publication). Published US Patent are https://patents.justia.com/patent/20240259351 and https://patents.justia.com/patent/20240203046.
Article on MilvusDB usage: https://www.linkedin.com/pulse/connecting-ibm-watsonxdata-milvus-attu-binoy-thomas-3jyyc/?trackingId=7dhiltyiQrCmgWXZ%2BQnSCw%3D%3D