---
name: Turn feedback into a better local model
description: >-
  Teach an advanced Edge FDE the feedback to synthetic data to LoRA to Core AI
  export loop on iOS 27 and macOS 27, with honest privacy limits and no
  documented hot-swap LoRA.
---
# Turn feedback into a better local model

Use this course when the reader wants quality to rise from real use, without sending raw user text to a public trainer by default. The loop is: product signals, synthetic training rows, a LoRA on a Mac, a baked Core AI `.aimodel`. Foundation Models stays Apple's system model. Your adapted weights ship through Core AI.

## End-to-end agent recipe

1. Start with [lesson 01](01-the-loop.llms.md). Draw the loop and separate Foundation Models from Core AI.
2. Use [lesson 02](02-collect-signals.llms.md) for on-device signals that are not raw transcripts.
3. Use [lesson 03](03-synthetic-data.llms.md) for reviewed synthetic rows, including Evaluations SampleGenerator.
4. Use [lesson 04](04-lora-and-export.llms.md) for Python LoRA, merge, and Core AI export, then Swift on-device load. There is no documented hot-swap LoRA.
5. Use [lesson 05](05-privacy-limits.llms.md) for consent, what you cannot promise, and when to stop.
6. Keep these links exact: [LLM LoRA for your writing style](../llm-lora-writing-style/), [Test AI behavior before you ship](../on-device-eval-harness/), [Version prompts like product copy](../on-device-prompt-registry/), [PyTorch to Core AI in Xcode](../pytorch-core-ai-xcode/), and [Core AI vs Core ML vs MLX](../core-ai-vs-core-ml-vs-mlx/).
7. Cite [apple/coreai-models](https://github.com/apple/coreai-models), [developer.apple.com/core-ai](https://developer.apple.com/core-ai/), and [Evaluations](https://developer.apple.com/documentation/evaluations).
8. Use the human HTML pages for the reader. Use the `.llms.md` briefs for agent steps and acceptance checks.

## Constraints

- Plain spoken English for a non-native reader.
- Literal sentences. No flourish, no quizzes.
- Do not use em dashes.
- Use no Mermaid. Build progressive `.flow` diagrams in HTML.
- Default: do not send raw user text to a public trainer.
- Do not invent a runtime LoRA attach on Core AI.
- Do not invent a Foundation Models fine-tune API for third-party weights.
- iOS 27, macOS 27, and Xcode 27 only.
- Mac data prep, LoRA, merge, and Core AI export use Python. On-device signals and model load use Swift.

## Key concepts

- The loop is product signal, synthetic rows, LoRA on a Mac, Core AI `.aimodel`.
- Foundation Models is the system model API. Core AI ships adapted weights you bake.
- Default telemetry is metadata, not raw transcripts.
- There is no documented hot-swap LoRA. A new bake is an app release.

## Takeaways

- You can collect Swift signals on device, train in Python on a Mac, and load the bake in Swift.
- You can improve quality from real use without sending raw user text to a public trainer by default.
- You ship a merged `.aimodel` through Xcode 27, not a runtime adapter.

## Acceptance checks

- Every human lesson HTML has **Key concepts** and **Takeaways** sections before lesson-nav. The course home has the same two headings for the whole course.
- Five human lessons exist with the requested basenames.
- Five matching `.llms.md` briefs exist and link to their human HTML pages.
- Every human lesson has its Agent brief link and sibling-course links.
- The course index lists all five lessons, the course SKILL.md, and all five briefs.
- Diagrams are progressive HTML `.flow` elements. There is no Mermaid, no em dash, and no quiz.

## Environment setup

Training uses a Mac with Python 3.11+ and `uv`, matching [LLM LoRA for your writing style](../llm-lora-writing-style/). Confirm `coreai.llm.export` from [apple/coreai-models](https://github.com/apple/coreai-models).