Sage Evolving Expert System - Cultivate Your AI Expert Advisor

Core Philosophy

Sage is atypica.AI's evolving AI expert system that makes AI truly "smarter over time" through three key mechanisms: memory documents, knowledge gap tracking, and supplementary interviews.

Core Value:

1. Memory as Expertise: Build expert capabilities based on structured knowledge documents (Memory Documents)
2. Proactive Learning: Discover knowledge blind spots through conversations and actively supplement the knowledge base
3. Continuous Evolution: Every conversation and interview makes the expert stronger
4. Traceability: Complete knowledge source tracking and version history management

Analogy:

• Traditional AI: One-off Q&A, forgets everything after the conversation ends
• Sage: Like cultivating a real human expert who continuously learns, remembers, and evolves

---

Overview: Sage vs Traditional AI

Real-World Case Comparison

Scenario: Consulting on company's UX design guidelines

Traditional AI (10 minutes):

Sage Expert (30 seconds):

Efficiency Gains:

• Time: 10 minutes → 30 seconds (95% time saved)
• Accuracy: Generic advice → Company-specific guidelines (100% accurate)
• Persistence: One-time → Permanent memory

---

How Does Sage Work?

Part 1: Create Expert in Three Steps

Step 1: Parse Knowledge Sources

Upload Professional Materials:

• File Upload: PDF, TXT, Markdown, Audio, DOCX (up to 10 files)
• Text Paste: Direct text input
• URL Import: Fetch content from web pages (using Jina Reader API)

Automatic Parsing:

---

Step 2: Extract Knowledge and Build Memory Document

AI Analysis:

Memory Document Example:

Create Version 1:

---

Step 3: Knowledge Analysis

AI Identifies Knowledge Gaps:

Knowledge Gaps Recorded:

---

Part 2: Proactive Learning Mechanism

Mechanism 1: Discover Knowledge Gaps During Conversations

Real Conversation Scenario:

Key Features:

1. Asynchronous Analysis: After conversation ends, uses Gemini 2.5 Flash for low-cost analysis
2. Automatic Identification: AI determines which questions weren't answered adequately
3. Link to Original Conversation: Can click to view complete conversation context
4. Continuous Optimization: Every conversation may uncover new knowledge gaps

---

Mechanism 2: Supplementary Interviews

Trigger Condition: When unresolved Knowledge Gaps exist

Automatic Workflow:

Step 1: Generate Interview Plan

Step 2: AI Interviewer Conducts Interview

Step 3: Automatically Update Knowledge Base

---

Part 3: Knowledge Evolution Tracking

Version Control

Memory Document Version History:

---

Knowledge Gap Lifecycle

Gap Status Flow:

Gap Detailed Information:

---

Use Cases

Scenario 1: Enterprise Knowledge Transfer

Background: A company's senior UX designer is about to retire and needs to transfer 20 years of experience.

Operation Flow:

Step 1: Create Expert

Step 2: Build Knowledge Base

Step 3: Discover Knowledge Gaps

Step 4: Supplementary Interview

Step 5: New Employee Use

Value:

• Knowledge Permanently Saved: After designer retires, knowledge remains accessible
• Consult Anytime: New employees not limited by senior staff availability
• Continuous Evolution: New employee questions continuously optimize Sage

---

Scenario 2: Personal Learning Assistant

Background: University student Wang is studying machine learning with lots of course materials and notes.

Operation Flow:

Step 1: Create Learning Assistant

Step 2: Learning Process

Step 3: Continuous Optimization

Value:

• Personalized Learning: Based on own notes and understanding, not generic answers
• Timely Q&A: No need to wait for office hours, ask anytime
• Fill Knowledge Gaps: Sage proactively discovers learning blind spots and reminds to supplement

---

Scenario 3: Industry Research Expert

Background: FinTech industry analyst needs to track industry dynamics and trends.

Operation Flow:

Step 1: Create Industry Expert

Step 2: Quick Insights

Step 3: Continuous Updates

Value:

• Quick Integration: Automatically integrates multiple reports without manual extraction
• Continuous Updates: Immediately updates after new reports released, stays current
• Discover Contradictions: Automatically discovers viewpoint differences across reports

---

Sage vs NotebookLM

Core Differences

Sage Emphasizes "Cultivating Experts":

• Knowledge is dynamically evolving
• System proactively discovers knowledge gaps
• Supplements knowledge through structured interviews
• Complete knowledge evolution tracking

NotebookLM Emphasizes "Understanding Materials":

• Knowledge is static (based on uploaded documents)
• Passively responds to user queries
• Innovative output formats (podcasts, video summaries)
• Quick onboarding, low usage threshold

---

Capability Boundaries

✅ What Sage Can Do

1. Knowledge Import

• ✅ PDF, TXT, Markdown, Audio, DOCX (up to 10 files)
• ✅ Text paste
• ✅ URL import (automatically fetch web content)

2. Knowledge Management

• ✅ Versioned memory documents (retain 20 versions)
• ✅ Knowledge source tracking
• ✅ Change history management

3. Proactive Learning

• ✅ Initial knowledge analysis (identify knowledge gaps)
• ✅ Discover knowledge gaps during conversations (asynchronous analysis)
• ✅ Supplementary interviews (AI automatically generates interview plans and questions)
• ✅ Automatically update knowledge base

4. Expert Consulting

• ✅ Public expert homepage
• ✅ AI-generated recommended questions
• ✅ Private conversations (visible only to owner)
• ✅ File attachment upload (as conversation context)

5. Traceability

• ✅ Complete knowledge source tracking
• ✅ Knowledge Gap lifecycle management
• ✅ Version control and change history

---

❌ What Sage Cannot Do

1. Real-time Data

• ❌ Does not support real-time data sources (e.g., real-time stock prices)
• ✅ Can periodically manually update knowledge sources

2. Cross-expert Collaboration

• ❌ Currently does not support multiple people jointly maintaining one expert
• ✅ Each user can create multiple independent experts

3. Automated Updates

• ❌ Will not automatically crawl latest information
• ✅ Incremental processing after user manually adds new knowledge sources

4. Tool Invocation

• ❌ Currently does not support web search, deep research, and other tools
• ✅ Planned for future support

---

Best Practices

1. Create High-Quality Experts

Provide Diverse Knowledge Sources:

Content Depth Recommendations:

• Overview Level: 10-20% (quickly understand the big picture)
• Detailed Level: 60-70% (deep professional knowledge)
• Case Level: 10-20% (real cases and decision processes)

---

2. Fully Utilize Supplementary Interviews

Create Interviews Promptly:

Interview Techniques:

---

3. Continuously Observe Conversation Quality

Regularly Review Gaps Tab:

---

4. Cost Optimization

Process Knowledge Sources in Batches:

Selective Interview Creation:

---

Technical Architecture

Data Model

Core Entity Relationships:

Memory Document Version Management:

• Each version automatically numbered (v1, v2, v3...)
• Structured documents in Markdown format
• Records change source (initial creation/interview supplement/manual edit)
• Detailed change descriptions
• Complete creation timestamp records

Knowledge Gap Traceability Information:

• Title and detailed description
• Severity level (Critical/Important/Nice-to-have)
• Current status (Pending/Resolved)
• Gap source (initial analysis/conversation discovery/system suggestion)
• Associated conversation records (if from dialogue)
• Resolution method records (via interview/manually marked)
• Associated interview records (if resolved via interview)

---

AI Model Strategy

Cost Optimization Design:

1. Manual Trigger: All three processing steps require manual user trigger, avoiding unexpected consumption
2. Asynchronous Processing: Conversation quality analysis and post-interview updates use background asynchronous processing
3. Lightweight Models: Non-critical tasks use lower-cost models (Gemini 2.5 Flash)
4. Incremental Updates: Supports incremental processing after adding new knowledge sources

---

Future Outlook

Near-term Improvements (within 3 months)

1. Version History UI

   • Visualize version history
   • Compare differences between versions
   • Rollback to historical versions

2. Real-time Processing Progress

   • WebSocket/SSE real-time updates
   • Processing progress bars
   • Detailed processing logs

3. Expert Public/Private Control

   • Choose whether to publish expert homepage
   • Set access permissions
   • Share expert with team members

Mid-term Improvements (within 6 months)

1. Tool Invocation Support

   • web search (real-time search for latest information)
   • deep research (deep research capability)
   • Custom MCP tools

2. Expert Collaboration

   • Multiple people jointly maintain one expert
   • Permission management (owner/editor/viewer)
   • Change review process

3. Knowledge Base Export

   • Export memory documents (Markdown/PDF)
   • Export complete knowledge base (JSON)
   • Migrate to other systems

---

Summary

Sage Evolving Expert System is an innovative feature of atypica.AI that achieves continuous evolution of AI experts through three key mechanisms: memory documents, knowledge gap tracking, and supplementary interviews.

Core Value

1. Continuous Evolution: Every conversation and interview makes the expert stronger
2. Proactive Learning: System proactively discovers knowledge gaps and supplements them
3. Knowledge Transfer: Make tacit knowledge explicit, permanently preserving expert experience
4. Traceability: Complete knowledge source and change history

Use Cases

✅ Suitable For:

• Enterprise knowledge transfer (senior employee experience transfer)
• Personal learning assistant (course note organization)
• Industry research expert (continuously track industry dynamics)
• Professional consulting (law, medicine, technology, and other fields)

❌ Not Suitable For:

• Real-time data queries (e.g., stock prices)
• General Q&A (ChatGPT is more suitable)
• One-time document understanding (NotebookLM is faster)

Relationship with Other Features

Feature Synergy:

• File Attachments: Upload files to conversation vs upload to Sage knowledge base, complementary
• Memory System: User-level memory vs expert-level memory, dual-layer architecture
• MCP Integration: Sage can call MCP tools in the future, expanding expert capabilities

---