6/12/25 - Boston Java Users ACM - Java & AI: A real world use case

Lightning talk from https://www.meetup.com/nejug1/events/306414600

Transcript

1. @edeandrea Java & AI A real world use case Eric Deandrea, Red Hat (soon to be IBM) Java Champion | Senior Principal Developer Advocate
2. @edeandrea @edeandrea The Problem
3. @edeandrea
4. @edeandrea
5. @edeandrea
6. @edeandrea
7. @edeandrea @edeandrea The Solution
8. @edeandrea
9. @edeandrea
10. @edeandrea
11. @edeandrea @edeandrea Non-functional requirements & constraints
12. @edeandrea
13. @edeandrea
14. @edeandrea @edeandrea Initial Design
15. @edeandrea
16. @edeandrea
17. @edeandrea @edeandrea Current Design & Architecture
18. @edeandrea Read images from filesystem Describe image Create .docx doc on filesystem Store .docx doc Add global view sharing perm & get URL Generate QR code Store QR codes on filesystem Add to manifest Store manifest as .xlsx
19. @edeandrea
20. @edeandrea Model Context Protocol (MCP) Instead of exposing tools from your code, discover and use remote services https://modelcontextprotocol.io
21. @edeandrea Read images from filesystem Describe image Create .docx doc on filesystem Store .docx doc Add global view sharing perm & get URL Generate QR code Store QR codes on filesystem Add to manifest Store manifest as .xlsx
22. @edeandrea
23. @edeandrea Read images from filesystem Describe image Create .docx doc on filesystem Store .docx doc Add global view sharing perm & get URL Generate QR code Store QR codes on filesystem Add to manifest Store manifest as .xlsx https://github.com/edeandrea/salem-historical-society-image-classifier
24. @edeandrea @edeandrea Lessons Being Learned
25. @edeandrea • The model matters • Tell the model what’s going on • Keep LLM arguments concise • Don't use JPA entities as tool arguments • Be (overly) descriptive • MCP is like the wild west • Testing is hard • Faster time to market without AI? Lessons learned (or in process of learning)
26. @edeandrea @edeandrea Thank you! Code Slides Recording