How I use LLMs to let overlooked aspects of my life emerge (danielfalbo.substack.com)

A former maker of an AI diary called Deeply outlines a practical workflow for using large language models to surface overlooked aspects of your life by analyzing personal journals. The author recommends exporting notes (Markdown/plain text), opening the directory with an LLM-enabled tool that has file-system/CLI access (e.g., Cursor, Claude), and priming the model with repo structure and key docs (README, timeline-of-my-life). By feeding sentence- and word-level content plus metadata like commit messages and journaling frequency, the model is asked to infer psychological patterns, vulnerabilities, and life context, then cross-check those inferences against the psychological literature and produce a realistic timeline of exercises and actions. Technically, the process emphasizes fine-grained prompts that push models to read structure and semantics (including single-word choices), use provenance data (timestamps/commits) for temporal insights, and generate actionable outputs (3-bullet psychological summaries, timelines, even LaTeX reports). Important practical notes: privacy-conscious setups are possible, consistent journaling amplifies value, and human vetting is essential (the author suggests “roast me” prompts to encourage blunt feedback). For AI/ML practitioners this highlights opportunities—modeling longitudinal personal data, leveraging metadata for behavioral inference, improved prompting strategies, and tool integrations—while underscoring ethical/privacy trade-offs and the greater value of sustained journaling.