AI w/o Spaghetti: Clean Architecture in the Age of AI
summary
Prototype scrapes legacy HTML, stores recipes in PostgreSQL, and adds OCR (Tesseract) plus translation so a chatbot can read and translate images; models used include OpenAI, Entropic CLI and a local LLM (Gemma) with a fallback routing demo that directs simple queries locally and advanced ones to the cloud. Experiments show 6 branches to fix a copy‑paste bug, 6–7 hour waits on cloud iterations, and that a Quarkus+Java backend plus OpenAI+Entropic+local Gemma multimodel setup works steadily. Enforced with ArcUnit architecture tests and a recommended multi‑module Gradle split; guardrails for token and content limits are essential.
AI-written summary · about
chapters (8)
- 0:00Introduction and presenters
- 1:39Experiment challenges and goals
- 6:14Legacy to modern web app demo
- 14:31Clean architecture explained
- 21:33Applying clean architecture
- 27:15Adding a chatbot and guardrails
- 36:42Key findings and lessons learned
- 42:19Q&A session
related talks
41mAI without spaghetti: Clean architecture in the age of AI
52mAI for Busy Java Developers
48mBuilding AI Powered Applications with Java
51mJava Meets AI: Empowering Developers to Build Intelligent Apps
27mBuilding AI applications: Successes and Lessons
45mWeb Development in the Age of AI
48mHow to Make Smart Architecture Decisions when Building Gen AI Apps
26mAI-first frontend architecture