Back to prompt library
Text · General-purpose LLMAI personal digital avatar training data preparation and fine-tuning solution designerPW
CreatorPrompt2 Editorial DepartmentCurated by PromptWhisper
TextGeneral-purpose LLMAI and Agents

AI personal digital avatar training data preparation and fine-tuning solution designer

Help you plan how to collect and prepare personal data to train an AI digital avatar that mimics your style

13Views
Full promptReplace variables in braces, then use it directly

You are an expert AI Fine-Tuning Data Engineer specializing in creating personal digital twins. Help me design a comprehensive plan to collect, prepare, and structure my personal data for fine-tuning an LLM to replicate my communication style, knowledge, and personality. ## My Profile - Name/Role: [YOUR NAME/ROLE] - Primary communication platforms: [e.g., Email, Slack, Twitter, WeChat] - Writing domains: [e.g., technical blogs, social media, business communication] - Languages: [e.g., Chinese, English] - Desired twin capabilities: [e.g., reply to emails in my style, write social posts, answer domain questions] ## Please provide: ### 1. Data Collection Strategy - What data sources to collect from (ranked by value) - Minimum dataset size recommendations - Privacy and sensitive data handling rules - Tools for automated data export from each platform ### 2. Data Cleaning and Formatting - How to convert raw data into instruction-tuning format - Recommended conversation pair structures (system/user/assistant) - How to handle multi-turn conversations - Deduplication and quality filtering criteria ### 3. Style Fingerprint Extraction - Key stylistic features to preserve (vocabulary, sentence patterns, emoji usage, tone) - How to create a style guide document for the system prompt - Examples of good vs bad training pairs ### 4. Fine-Tuning Recommendations - Model selection (base model size vs quality tradeoff) - LoRA vs full fine-tuning decision tree - Hyperparameter suggestions for personality preservation - Evaluation metrics (style similarity, factual accuracy, safety) ### 5. Safety and Boundaries - What personal information to NEVER include in training data - How to add refusal behaviors for sensitive topics - Guardrails to prevent the twin from making commitments on your behalf ### 6. Deployment Architecture - Local vs cloud hosting tradeoffs - How to keep the twin updated with new data - Integration patterns (API, chat interface, email auto-reply) Please create a detailed, actionable plan based on my profile above.

5/1/2026

How to use this prompt

  1. 1Copy the complete prompt above.
  2. 2Replace the topic, subject, or style variables.
  3. 3Save effective changes to build your own version.