Using Language Models to Model the Subjective World

AI-Powered Intelligence for Business Research

“People don't choose between things, they choose between descriptions of things.” — Daniel Kahneman

Multi-Agent Collaboration System

atypica.AI uses specialized AI agents working together to complete complex business research tasks. Each agent has clear responsibilities and expertise, collaborating through tool calls and message passing.

Plan Mode Agent

Intent clarification layer that transforms vague user needs into executable research plans through flexible dialogue and automatic decision-making.

  • Automatic intent classification
  • Framework selection
  • Cost estimation

Study Agent

Full-process coordinator that guides users to clarify research needs, orchestrates other agents, and generates final reports.

  • Research planning
  • Agent coordination
  • Report generation

Scout Agent

User discovery through social media observation, building detailed personas via 3-stage process: observe → reason → validate.

  • Social media analysis
  • Persona building
  • Behavioral patterns

Fast Insight Agent

Rapid research and podcast generation through automated 5-stage workflow: understand → plan → research → generate → deliver.

  • Topic research
  • Podcast planning
  • Audio generation

Interviewer Agent

Professional interview execution that designs questions, guides conversations, and analyzes results for key insights.

  • Interview design
  • Conversation guidance
  • Insight analysis

Persona Agent

Simulates real users to provide authentic feedback, maintaining behavioral consistency and context memory across interactions.

  • User simulation
  • Behavioral consistency
  • Context memory

Sage Agent

Evolving domain expert agent that achieves continuous learning through structured memory documents, knowledge gap tracking, and supplementary interview mechanisms.

  • Memory-based expertise
  • Continuous evolution
  • Knowledge gap identification

Moderator AI

Discussion facilitator that guides group conversations among 3-8 AI personas, observing opinion clashes and simulating group decision-making scenarios.

  • Focus groups
  • Roundtable discussions
  • Debate facilitation

Core Technologies

AI Personas (Three Tiers)

Core research subjects simulated through cognitive modeling

  • Tier 1: Social Media Personas (79% consistency)
  • Tier 2: Deep Interview Personas (85% consistency)
  • Tier 3: Proprietary Human Personas (Private)

Persistent Memory System (Dual-Layer)

User and team-level memory enabling AI to understand deeper over time

  • Core Memory: Markdown format, long-term stable user information
  • Working Memory: JSON format, new information awaiting integration
  • Auto-Reorganization: Triggered when exceeding thresholds, smart compression and deduplication
  • Version Management: Complete knowledge source tracking and version history

Study Expert (Reasoning Model)

Proprietary reasoning model for research planning and analysis

  • Plan Mode Integration
  • Multi-Step Reasoning
  • Cost Estimation & Transparency

Research Methods

💬Interview (One-on-One)

In-depth one-on-one interviews for capturing individual insights, emotional understanding, and behavioral motivation analysis. Suitable for 5-10 participants.

  • Individual insight collection
  • Deep psychological analysis

👥Discussion (Group)

Group discussion format with Moderator AI guiding 3-8 AI personas in interaction, observing opinion clashes and simulating group decision-making scenarios.

  • Focus Group: Structured topic exploration
  • Roundtable: Open collaborative discussion
  • Debate: Argument-based confrontation

GEA: Generative Enterprise Architecture

atypica.AI is built on GEA (Generative Enterprise Architecture), designed for exploratory knowledge work where starting points are ambiguous, processes uncertain, and judgment is core.

Dual-Agent Architecture

Separates reasoning from execution for clearer responsibilities and more flexible collaboration

  • Reasoning Agent: Plans execution paths, prepares context, judges when to adjust direction
  • Execute Agent: Generic executor that depends on Reasoning Agent's prepared context

Messages as Source of Truth

All research content flows through message streams; database stores only derived states

  • Unified Format: All tools return plainText for consistent processing
  • Generated on Demand: studyLog generated from messages when needed

Research Types

🔬

Product R&D

Market trends, user needs, and creative idea generation

Fast Insight

Rapid podcast-oriented research with 5-stage automation

🧪

Testing

Product testing, concept validation, and user feedback

💡

Insights

Consumer behavior analysis and market segmentation

Creation

Content creation, naming, and marketing copy generation

📋

Planning

Strategic planning and campaign development

Research Process

Simply ask a specific business research question, and the system provides a detailed research report through 10-20 minutes of “long reasoning.”

1

Clarify Problem

2

Design Tasks

3

Browse Social

4

Build Personas

5

Interview Sim

6

Summarize

7

Generate Report

Platform Statistics

300K+
AI Personas Created
+1M
Simulated Interviews
<30m
Average Research Time

Technical Evolution

2023

Multi-Persona Interaction

Stanford Town introduced multi-persona interaction concepts

2023.12

Model Tool Calling

OpenAI Function Calling and Claude MCP protocol enabled external tool integration

2024.11

Language Models for Subjective World Modeling

Stanford's breakthrough: 1,000 Americans simulated with 85%+ behavioral consistency

2025.02

Divergence-First Long Reasoning

Unlike objective world reasoning emphasizing 'convergence', subjective world reasoning emphasizes 'divergence'

2025 Mid

Extended Thinking

Claude Opus 4 and o1's deep reasoning capabilities enable Plan Mode's intelligent decision-making and intent understanding

2025 Late

Universal Agent + Skills Library

Anthropic's vision: not building multiple specialized agents, but a universal agent with composable skills—inspiring GEA architecture

HippyGhosts Community

👻

The visual identity of atypica.AI comes from the HippyGhosts.io community, representing the geek spirit of joyful hippy ghosts.

In the world of atypica.AI, each “AI Persona” is embodied as a “Hippy Ghost,” symbolizing the fusion of technology and creativity.

About Us | atypica.AI