Flutter Apps in the New Age of Non-Deterministic Data
summary
Hope Street Group builds a Flutter app using LLMs (CHBT3.5) to generate career “professionals” and launches in four schools; generation improves from ~60–70% valid JSON to ~90–99% after adding function calling and tuned temperature/top-p. Pipeline uses prompt engineering, JSON validation, sanitation, type coercion and defaults (age => 25) so UI never crashes, plus moderation via the moderation API and a custom banned-title list; embeddings with cosine similarity are planned to catch slang and misspellings.
AI-written summary · about
chapters (9)
- 0:00Non-deterministic data in Flutter
- 1:53Hope Street Group’s career app
- 7:11LLM-generated data challenges
- 10:01Prompting and API tuning
- 13:18Function calling for structured data
- 14:43Validation and error handling
- 17:31Sanitization and defaults
- 18:34Content moderation
- 20:17UI for dynamic data
related talks
36mFlutter-Google's latest innovation for mobile, web, and desktop apps
22mArchitecture First: Foundations for Long Term Success
20mHow Flutter Can Change the Future of Urban Communities
13mBuilding platform adaptive apps
46mHow to Cook Flutter Properly
44mMultimodule Flutter App
30mFlutter at scale-experience from 25+ Flutter Devs working together on a mobile banking app
45mBuilding Products Full-Stack with Dart and Flutter