ammonhaggerty/my-digital-twin

Implements a curated instruction prompt (~30,000 words) built from articles, projects, and personal narratives—preprocessed through Claude and GPT o1 into modular summaries to overcome context window limitations. Tested across local LLMs (Llama 3), Claude 3.5 Sonnet, and ultimately OpenAI's 128k+ context models, settling on prompt engineering techniques like hierarchical markdown formatting and synthetic narrative summaries to enable consistent persona representation. The project explores authentic AI representation that avoids both the uncanny valley of realistic cloning and "agent identity entanglement," positioning the digital twin as an autonomous proxy with agency rather than a simple chatbot interface.