AI/MLDevelopersDevOpsLet's TalkVideo

Crawl-Walk-Run is the right strategy to build GenAI applications: Roman Kharkovski

In this episode of Let’s Talk about AI, we discuss the importance of establishing a robust platform for building generative AI (GenAI) applications and cover key elements, such as data management, model selection, scalability, cost, and system integration. Roman Kharkovski, Principal Architect at Qarik, explains the importance of understanding the business needs before proceeding with the development of an AI application.
Swapnil BhartiyaBy 409 views
0

In this episode of Let’s Talk about AI,  we discuss the importance of establishing a robust platform for building generative AI (GenAI) applications and cover key elements, such as data management, model selection, scalability, cost, and system integration. Roman Kharkovski, Principal Architect at Qarik, explains the importance of understanding the business needs before proceeding with the development of an AI application. We chat about the spectrum of degrees of automation for business processes from co-pilot to auto-pilot and the tactics and strategy when implementing GenAI applications.

Highlights: 

Implementing AI in business processes

  • Kharkovski shares his experience with GenAI, projects in the area, and the shift in focus from DevOps to business use cases.
  • Gemini is used for creative writing, co-pilot work, industry-specific solutions, and process improvement in various industries, with a focus on financial services.
  • Roman Kharkovski recommends a crawl-walk-run approach for automating business processes, starting with copilot phase and moving to autopilot for complete automation.

Building scalable AI systems for ingesting various data formats

  • Kharkovski discusses the challenges of building a scalable and resilient AI system for ingesting and processing various types of data.

Cost-effective AI solutions for business workflows

  • Kharkovski emphasizes the importance of fine-tuning embeddings models for domain-specific understanding and using a combination of AI and software engineering approaches to solve problems.
  • Cost of using GenAI can be high, but optimizing implementation can reduce it by two orders of magnitude.

Building AI models and deploying them quickly

  • Kharkovski shared lessons learned from building Gemini projects for 9 months, including the importance of a robust data ingestion system and reusable frameworks for future engagements.
  • The accelerator built by Kharkovski’s team separates into four tiers: data ingestion, query engine, UI, and shared services for cost control and security.

Guest: Roman Kharkovski (LinkedIn)
 Company: Qarik Group (Twitter)
 Show: Let’s Talk

You may also like