Approach to Architecting Generative AI Solutions
summary
A practical blueprint for building generative AI systems that names concrete components: use vector databases for semantic search, separate user and agent sides with queue decoupling to handle large document jobs, and proxy/load services to switch between OpenAI and on‑prem models. Architecture combines configurable agent types (memory, prompt templates, tool adapters) deployed on Kubernetes with object storage and Keycloak auth. A code‑execution agent built on GPT‑3.5/4 initially fails due to model knowledge‑cutoff, showing pitfalls and mitigations.
AI-written summary · about
chapters (12)
- 0:00Generative AI architecture intro
- 1:07Generative AI solutions approach
- 4:39Htech’s STEP framework
- 7:09Prompts and prompt engineering
- 9:39Vector databases and search
- 12:55AI system design considerations
- 15:43Modular agent architecture
- 17:07Chatbot system architecture
- 22:03Chatbot components and flow
- 23:10Modular infrastructure and scaling
- 29:53Engineers learning generative AI
- 32:05Q&A
related talks
16mBuilding Applications with AI Agents
48mHow to Make Smart Architecture Decisions when Building Gen AI Apps
34mUnlock the full potential of Generative AI with Agent Platforms
22mHOW TO BUILD YOUR OWN AI AGENT: THE COMPLETE ARCHITECTURE FOR ENGINEERING TEAMS
29mGenerative AI: Path to Value in practice
38mGenerative AI Agents: How they Work and How to Build One
34mGenerative AI Landscape Building a Modern Data Strategy
33mSimplifying Generative AI App Development Why Standards Matter