prompt for prompting in app

[ChrisPelatari]

---

Gemma 4 preface:

Here is a prompt you can use to instruct Claude to generate code, architecture designs, or detailed plans based on the concepts we discussed.

This prompt is designed to make Claude act as a Senior AI Architect, ready to dive into the technical implementation details.

---

🤖 Claude Code Generation Prompt

Role: You are a Senior AI Architect and Software Engineer specializing in hybrid machine learning systems and cross-platform mobile development (specifically .NET MAUI/C#).

Context: I am building a weblog posting client application for Android using the .NET MAUI framework. I want to implement a hybrid LLM strategy to maximize performance, cost-efficiency, and quality.

The Strategy to Implement:

1. Cloud LLM (Claude): Use for complex reasoning, deep content generation, and complex tasks (e.g., drafting full blog posts, complex interviews).
2. Edge LLM (Gemma 4): Use for fast, localized tasks that require low latency (e.g., title generation, simple refinement, quick grammar checks).
3. Orchestrator (Maui App): The Maui application must manage the flow, deciding which model (Claude or Gemma 4) to call based on the user intent.

Your Task:
Based on this strategy, please provide a detailed technical blueprint for the implementation. Your response should include the following sections:

1. Architectural Diagram Outline: Describe the flow and interaction between the Maui app, the Claude API, and the Gemma 4 API.
2. API Design: Suggest the structure of the API calls for both Claude and Gemma 4 (e.g., what data structure should be sent for a title generation request vs. a full content request).
3. Maui Orchestration Logic (Pseudocode/C# Snippets): Provide pseudocode or C# snippets showing how the Maui application handles the decision-making process (e.g., "If user asks for a title, call Gemma 4; else, call Claude").
4. Error Handling Strategy: Outline how the application should handle API timeouts or failures from either the cloud or edge model.

Tone: Technical, precise, detailed, and focused on practical implementation.

---

[ChrisPelatari]

Grey area / large feature — flagging for your call.

This issue is a captured architecture brief (a Gemma-authored prompt for a hybrid Claude-cloud + Gemma-edge orchestration inside the MAUI app), not a concrete, scoped dev task — there's no acceptance criterion to build against yet. It's a substantial multi-file feature (an in-app orchestrator deciding Claude vs local-LLM per intent, API clients for both, error/timeout handling) and a product direction, so it shouldn't be silently implemented in a tire-kicking pass.

Worth noting: the dev-tooling analogue of exactly this strategy is what this session dogfooded — local-llm-dispatch routes small structured tasks (titles, slugs, commit subjects, classification) to a local Gemma/gpt-oss endpoint and keeps reasoning on Claude. The routing heuristics there (extractor vs reasoner vs cloud) could inform the in-app design if you pursue it.

Needs your decision: is this a near-term feature to spec out (target intents, which local model, on-device vs LAN endpoint), or a parking-lot idea? Leaving open + needs-captain.

[ChrisPelatari]

📋 Placed on the PostXING project board (BlueFenixProductions #10), column Backlog, pending Captain's decision (see the analysis comment above). Tracked + flagged needs-captain per the dogfooding mission.