The Future of APP Development
summary
Builds a Dioxus-specific copilot by fine-tuning open models (LoRA on Llama family) with a synthetic dataset: ~8,000 format-following generations produced 1,500 high-quality examples used for tuning, raising usable output from ~10% to ~20%. Uses LoRA fine‑tuning, constraint/structure generation and a local runtime called cisum to run quantized models on commodity hardware (MacBook / T4/A100). Ships an agent that auto-tests UI in a headless browser, generates responsive RSX/Tailwind cards live, and the demo agent successfully flags a swapped-content bug.
AI-written summary · about
chapters (13)
- 0:00Introduction to Dioxis
- 0:37Dioxis framework overview
- 3:04Challenges and goals
- 4:56AI in development
- 6:35The Dioxis co-pilot
- 9:10Leveraging local models
- 12:40Synthetic data and tuning
- 15:52Constraint generation
- 18:15UI generation demo
- 20:25Agentic verification
- 22:50In-place code refactoring
- 25:44Future roadmap
- 28:25Q&A session
related talks
37mEmpowering Developers with local AI
48mMastering the AI Shift: Adaptive Strategies for Software Evolution
54mKeynote: AI-Powered App Development
36mPanel Discussion
25mWeights vs. Words in Fine-Tuning & Enhancing Agentic AI Workflows
47mOwn Your LLM: Running Generative and Agentic AI Locally
21mFrom 46% to 90%: Fine-Tuning Tiny LLMs for On-Device Agents
27mQ&A #2