Quality Gate for AI: How We Built an AI Control Architecture [ukr]
This talk presents an architectural approach to AI quality assurance in which a second AI system plays the role of a “judge,” replacing traditional testing methods. Using our real-world AI Calories Tracker an application that recognizes 5,000+ food photos every day, we will demonstrate how this unconventional strategy lets us manage risk effectively and boost trust in artificial intelligence.
- What to expect:
- An analysis of why classical QA techniques fall short for AI systems
- A step-by-step look at how one AI model automatically assesses the performance of another
- A walkthrough of the quality-control workflow at every stage of the solution’s life cycle, from selecting the optimal model and tuning prompts to routine model updates in production
- Key takeaways on the real advantages this approach delivered, as well as the challenges we faced
Dmytro Demianov
Solution Architect, BetterMe
- Dmytro has been in IT for 8 years, working his way up from developer to architect
- Specializes in AI/ML and Cloud Architecture, designing AI solutions for business using AWS
- Believes that Gen AI works best when integrated into business processes and solves specific company tasks
- He is fond of technical literature, architecture and management, and recovers in nature
- Main principle: architecture should always create real value for business