Obsidian Metadata
| categories | Build |
| description | An App that reimagines the concept of the second brain in the age of AI. |
Trigger / Inspiration
1. Project Overview
Objective: To build an automated “Second Brain” that allows for frictionless capture of thoughts, tasks, and information, using AI to classify, store, and proactively surface relevant data. Core Philosophy: Reduce the human’s job to one reliable behavior (Capture) and let AI handle the cognitive load of organizing.
2. Technical Stack
- Interface (Capture & Surfacing): Slack (one private channel named
#sb-inbox). - Logic & Automation: Zapier or Make.com.
- Intelligence: Claude 3.5/GPT-4o (via API).
- Memory (Storage): Notion (Relational Databases).
3. System Architecture & Building Blocks
A. The “Dropbox” (Capture Point)
- Requirement:
A single Slack channel where every message is a new entry.An Audio first / link share kind of input. - User Flow: User talsk/types/pastes a thought into a user friendly omnipresent action. No tagging or sorting required.
B. The “Sorter” (AI Classifier)
-
Requirement: An AI prompt that acts as an API. It must accept raw text and return structured JSON.
-
Categories:
- People: Networking, contact details, follow-ups.
- Projects: Active goals with specific next actions.
- Ideas: Insights, creative thoughts, “someday” items.
- Admin: Errands, logistics, dates.
-
Logic: The AI must provide a Confidence Score (0.0 to 1.0).
C. The “Filing Cabinet” (Notion Databases)
Create 5 specific databases in Notion:
-
People:
[Name, Context, Follow-ups, Last Touched, Tags] -
Projects:
[Project Name, Status (Active/Waiting/Blocked), Next Action, Notes] -
Ideas:
[Name, One-liner Summary, Elaborated Notes, Tags] -
Admin:
[Task Name, Due Date, Status] -
Inbox Log (The “Receipt”):
[Original Text, AI Destination, Confidence Score, Timestamp]
D. The “Bouncer” (Guardrails)
-
Confidence Threshold: Set to 0.6.
-
Action: If AI confidence is
< 0.6, the system must not file the entry. Instead, it logs it in theInbox Logas “Needs Review” and pings the user in Slack for clarification.
4. Automation Workflows (The “Loops”)
Loop 1: Real-time Filing
-
Trigger: New message in Slack
#sb-inbox. -
Action: Send text to AI with a System Prompt: “Classify this into People, Projects, Ideas, or Admin. Return JSON only with fields: category, name, next_action, summary, confidence_score.”
-
Branching Logic:
-
If Confidence > 0.6: Create a page in the corresponding Notion database.
-
If Confidence < 0.6: Log in
Inbox Logand send Slack reply: “I’m not sure where this goes. Please clarify if this is a Person, Project, or Idea.”
-
-
Confirmation: Post a Slack reply to the original message: “Filed as [Category] in [Database].”
Loop 2: The “Tap on the Shoulder” (Daily Digest)
-
Trigger: Scheduled (e.g., 8:00 AM daily).
-
Action: Query Notion for “Active Projects” and “People to Follow Up.”
-
AI Step: Summarize findings into a < 150-word Slack DM.
-
Format:
-
Top 3 actions for today.
-
One thing you might be avoiding (stagnant project).
-
One small win from yesterday.
-
Loop 3: Weekly Review
-
Trigger: Sunday at 4:00 PM.
-
Action: Query all entries from the past 7 days in the
Inbox Log. -
AI Step: Generate a < 250-word summary of open loops, recurring themes, and 3 suggested actions for next week.
5. Correction Mechanism (The “Fix Button”)
-
Requirement: If the user replies to an automated filing confirmation with
fix: [correction], a separate automation should trigger. -
Logic: The AI parses the correction (e.g., “This should be a project, not an idea”) and moves the Notion entry to the correct database.
6. Prompt Engineering Guidelines for AI Studio
To one-shot the prompts, provide the AI with these constraints:
-
Reliability Over Creativity: Always return valid JSON. No conversational filler.
-
Action-Oriented: When processing projects, always extract a “Next Action” (e.g., “Email Sarah,” not “Work on website”).
-
Conciseness: Daily and weekly summaries must be mobile-friendly and fit on one screen.
Considerations
- Why not building it within Obsidian (maybe a plugin )

