new file mode 100644
index 0000000..825e4e5
@@ -0,0 +1,27 @@
+---
+type: idea
+title: me model
+tags: [ai, personalization, data, finetuning]
+status: raw
+first_captured: 2025-12-22
+updated: 2026-04-10
+sources:
+ - sources/ideaflow/2025-12-22_me-model---data-sources-imessages-email-discord.md
+ - sources/apple-notes/archived/potential-projects-archived.md
+ - sources/google-sheets-ideas.md
+---
+
+# me model
+
+a fine-tuned language model trained on your personal data: iMessages, email, Discord, notes/Obsidian, GitHub. the goal is an AI that genuinely understands you — your communication style, your knowledge, your projects, your relationships.
+
+Fireworks AI was identified for the finetuning infrastructure. this is the foundation layer that several other ideas build on: [[oncue|OnCue]] needs your professional history, the [[always-on-ai-assistant|always-on AI assistant]] needs your context, the [[personalized-autocomplete|personalized autocomplete]] needs your writing patterns, and the [[ai-conversationalist|AI conversationalist]] needs your speaking style. the [[life-search|life search]] idea is essentially the query interface for a me model.
+
+**spreadsheet evaluation:** originality 7/10, excitement 5/10, MVP 4-8 weeks. competitive landscape: Character.AI lets others create chatbots, Delphi.ai trains on specific people, Personal.ai builds from your data, Inflection Pi learns over time — none pull from iMessages + email + Discord + Obsidian + GitHub as a unified training corpus. data extraction is the hard part (iMessage needs local SQLite access, Discord export is limited, email needs OAuth). privacy implications significant. tech depth 8/10, labeled [STRONG] ("fine-tuning, data pipelines from iMessage/Discord/GitHub, eval — technically meaty but demo is weird: 'look, it talks like me'").
+
+---
+
+## timeline
+
+- [2025-12-22] captured in ideaflow — data sources and finetuning approach
+- [2026-04-10] google sheets evaluation — 7/10 originality, 5/10 excitement, MVP 4-8 weeks, [STRONG] tier