Linear Capital portfolio company Tezign launches atypica.AI, using "language models" to model the "subjective world"

线性资本·April 8, 2025

To Build & Create

If physics models the "objective world,"

then language models have the opportunity to model the "subjective world."

atypica.AI captures human decision-making mechanisms that data analysis cannot handle, providing deep insights for personal and business decision-making problems.

— Fan Ling, Founder and CEO of Tezign

Video | atypica.AI

Human decision-making is not entirely based on rational judgment; it is shaped by narrative logic, emotional preferences, and cognitive biases. Understanding these core elements that influence decisions has become a critical topic in business research.

Built on this insight, the business research agent atypica.AI was born:

  • It constructs "user agents" to "simulate" consumer personalities and cognition;
  • Through "interviews" between "expert agents" and "user agents," it analyzes consumer behavior and decisions, generating reports.

Compared to the flat labels of traditional data statistics, this system strives to reconstruct three-dimensional, humanized decision-making minds in virtual scenarios, making hidden consumer psychology visible as analyzable chains of logic. Through natural language processing, atypica.AI effectively identifies decision-making characteristics that traditional data analysis methods struggle to quantify, delivering deep behavioral insights for individual users and enterprise clients.

Click the image below to learn more about atypica.AI's product thinking and technical details 👇

Once logged into atypica.AI, users simply pose a specific business research question. The platform then delivers a detailed research report through 10–20 minutes of "long reasoning."

During this 10–20 minute reasoning process, atypica.AI automatically performs the following tasks:

  1. atypica.AI asks you 1–5 follow-up questions to clarify the research problem;

  2. atypica.AI designs a series of work tasks;

  3. atypica.AI browses social media according to these tasks (currently only Xiaohongshu, with richer sources to come);

  4. atypica.AI builds multiple "user agents" based on browsing results;

  5. atypica.AI interviews these "user agents," and if answers are insufficient, continues building additional agents;

  6. atypica.AI summarizes interview results;

  7. atypica.AI generates a report in the specified visual style;

"Nerd Stats" records metrics such as time spent, steps taken, number of agent roles, and tokens consumed during the process — serving as a kind of "Proof of Work" for the agent.

Note: atypica.AI's nerd stats

Testing, insight, planning, and co-creation are four typical scenarios in business research. Below are some examples:

1. Testing: Fast, low-cost testing to gather consumer feedback

Which topic for a Logitech mouse post on Xiaohongshu would perform better?

a. "Light as a feather, fierce as a tiger: my efficient workdays with the Logitech MX Keys Mini"

b. From keys to chips: how Logitech's silent technology works

c. One-click multi-device switching: how I doubled my productivity with Logitech FLOW

d. 30-day battery life is no dream: the energy-saving tech behind Logitech keyboards

e. The story behind ergonomic design: how the Logitech ERGO K860 saved me from carpal tunnel

2. Insight: Gaining open-ended consumer needs

I am the General Manager of LV (Louis Vuitton) Shanghai. What feedback do customers have about the in-store experience at our Shanghai LV boutiques? What areas need improvement, and what aspects are working well that we should continue to enhance? Please give me a holistic report.

3. Co-create: Developing new business ideas together with consumers

Co-create new product ideas for Mars's Crispy Rice Chocolate with young parents in tier-one cities?

4. Planning: Business planning aligned with market demand

INAH Non-Alcoholic Grape Beverage Marketing Plan

Additionally, atypica.AI can assist with personal decisions, such as:

1. Open-ended: Choosing the right Chinese restaurant for a birthday dinner?

2. Choice-based: How should I choose a portable monitor?

3. Planning-based: I'm a competitive swimmer — how should I plan for high school in the US or UK?

1. Subjective World Modeling Breakthrough: In 2024, a Stanford team achieved 85% accuracy in simulating human behavior through a thousand-agent experiment. Building on this, atypica.AI constructs an agent system that uses language models to reconstruct the emotional narratives and cognitive logic in consumer decision-making, breaking through the limitations of traditional tag-based user personas.

2. Tool-Calling Capability Upgrade: Integrating GPT-4 function calling with the Claude MCP protocol, the system enables agents to autonomously operate external tools. It can scan content platforms like Xiaohongshu in real time, completing trend prediction, sentiment analysis, and data traceability — pushing business research from responsive to exploratory.

3. Divergent Reasoning Architecture Innovation: Inspired by DeepSeek R1, it designs a "Creative Reasoning" framework centered on three pillars: case-based learning, inspiration stimulation, and feedback iteration. This simulates the unstructured thinking paths in human decision-making, supporting open-ended business problem exploration.

4. Multi-Agent Collaboration Implementation: Drawing on Manus's process visualization philosophy, it transforms interaction traces between user agents and expert agents into traceable research logs, making AI decision pathways transparent and enhancing brands' intuitive understanding of complex consumer psychology.

1. Individual/Team Version

  • Free quota: 3 complete research sessions
  • Continue using: leave a tip, buy us a coffee, thanks for the support ☕️

2. Enterprise Version (coming soon)

  • Deep drill-down: multi-layer follow-up questioning on report conclusions
  • Custom analysis: support for private data integration and customized report templates
  • Stay tuned

atypica.AI is now in open beta. Register via the official website to experience it (mobile supports background running; desktop offers a better experience). Access link (or click "Read Original" to go directly): https://atypica.ai/

About Linear Capital

Linear Capital is an early-stage investment firm focused on "frontier technology + industry," targeting frontier technologies such as data intelligence, digital new infrastructure, next-generation robotics, and new technology transformations in traditional sectors (biomedicine, materials, energy, etc.). These technologies are applied across vertical industries to dramatically improve efficiency, solve pain points, and enable industrial upgrading — generating outsized commercial returns through substantial industrial value creation. The firm currently manages ten funds with approximately $2 billion in total AUM.

Our investment stage focuses primarily on leading angel to Series A rounds, with individual investments ranging from $1 million to $10 million (or RMB equivalent).

To date, Linear Capital has made early-stage investments in over 120 teams including Horizon Robotics, Kujiale, Sensors Data, Tezign, Rokid, Guandata, and Agile Robots. The combined valuation of Linear's portfolio companies is approximately $20 billion.

In the near term, Linear Capital is working to become the premier "Data Intelligence Technology Fund," with a long-term vision of evolving into the most influential "Frontier Technology Application Fund."