Integrated operations: multi-AI multi-character emotional interactive drama + AI relationship evaluation + an offline interactive club
A brand-new species born from the strategic fusion of two seasoned teams, each with twenty years of deep expertise in its own field.
Twenty years deep in enterprise markets, with core executives holding proven IPO capital-operation track records. Independently developed and deployed GEO-SaaS ecosystems across six vertical industries and 11 AI-optimized GEO-MaaS intelligent engines. Four global hubs working in concert: Hangzhou · Hong Kong · Dubai · United States.
A veteran player in China's vertical relationship social sector, having lived through three full industry generations—the PC relationship era, the mobile social era, and the AI-powered social era. Deeply versed in the full chain of vertical social user psychology, risk-control systems, compliant operations, and user growth.
HKU & Fudan IMBA · Beijing Normal University (Psychology) · 10 years across venture capital, media and psychology
Trepanning across VC and media for a decade, Misue kept asking one question: what really lets one person be understood by another. The training in psychology set three lines MulanXu refuses to cross — no self-report questionnaires (people lie, behavior doesn't), no horoscope-style inference (no personality in a birthday), and no clinical judgments (boundaries themselves are professionalism). So the assessment is hidden inside a drama: 151 choices are 151 behavior samples written by your own hand. MBTI asks how you think; astrology sees when you were born; MulanXu watches what you do — this is not copy, it is method.
All 12 engine modules are developed in-house (no third-party APIs), and the technology can empower external partners.
From semantic understanding to emotion recognition, our emotional engine is fully controllable end-to-end.
A self-owned headquarters building plus a full-format offline club—while every competitor in the market runs a pure-online, asset-light model.
The world's first emotional matching system deeply fusing personality, interests, and individual traits with AI—a cultural moat that cannot be replicated.
The industry's only pay-as-you-go Token model—minimal user resistance, healthy repurchase rates, and decreasing marginal costs.
Continuously compliant, auditable, traceable, and risk-controlled. Zero facial-recognition abuse, with end-to-end encryption protecting emotional data.
General-purpose LLMs can chat, but they won't remember every hesitation across your 151 choices. Our Deduction Layer is something competitors cannot replicate by calling an API.
M01 Emotional Semantic Parsing—reading the emotion between the lines; M02 Multimodal Emotion Recognition—voice, expressions and rhythm; M03 Micro-Decision Intent Inference—inferring motivation from choices (the technical basis of "behavior observation").
M04 Long-Term Contextual Memory—remembers what you chose last time and why; M05 Behavioral Vector Profiling—151 choices converge into one computable line
M06 Character Personality Consistency—7 AI companions never break character (the source of the 95%+ consistency); M07 Plot Branch Generation—151 branch points unfolding in real time; M08 Narrative Style Transfer—the linguistic feel of the Ballad of Mulan.
M09 Mulan Personality Inference—behavioral vectors → 16 types × 5 dimensions (with confidence levels); M10 Relationship Tension Deduction—dual-vector difference → 136 relationship reports; M11 Personalized Content Generation—a bespoke daily Mulan post for every user.
M12 Compliance & Content Safety—fully auditable and traceable across the pipeline; emotional data never leaves the domain.
General-purpose LLMs lack long-term memory across 151 choices—personas drift;
General-purpose LLMs lack the "behavior observation → personality vector" modeling that is the very foundation of the Mulan assessment;
Emotional data is the most sensitive class of data—it must be owned, auditable and traceable.
We build on domestic base LLMs, and self-develop the "human-understanding" engine layer: personality inference, relationship deduction, daily Mulan posts—all self-built, all auditable.
The engine can empower external partners (mulanxu.net exposes capability APIs, not model weights). Competitors can buy LLM APIs, but they cannot buy four years of accumulated behavioral data or our proprietary deduction modeling—data and modeling are the real moat.
The Mulan Assessment did not invent its format from nothing. It translates three movements in Mulan's story — a drill before setting out, many costly rounds in the field, and merit recorded on return — into a method a team can observe on site, review, and use to rewrite the next plan.
After deciding to take her father's place, Mulan prepared bow, horse, weapons and clothes — then asked to drill a little first. The assessment borrows that move: put judgment into low-stakes situations and watch what a person chooses, not how they describe themselves.
Read the source & notes →The stele places Mulan in a long, continuous run of costly contests, not a one-off statement. The offline tabletop does the same: no model answer — division of work, decisions and reviews surface under real pressure.
Read the stele & notes →Return is not the end: merit is recorded and experience becomes the basis for the next campaign. The assessment ends not with a label, but with an action plan the team can take away.
Read the Ballad of Mulan →The online quick test takes about five minutes and thirty situational questions, returning a preliminary Mulan personality card and five indices: responsibility, courage, connection, wisdom and growth. It reads choice patterns under tension, not adjective-based self-description.
The team tabletop puts those tendencies into real division of work, decisions and review. Who verifies first when information conflicts, who steps up in a deadlock, who makes the handover explicit — all of it is visible on site.
A review is not a critique session. The team writes what it saw into a next-round action plan: who verifies, who hands over, under what conditions to switch — then tests whether the change helped.
If you know MBTI, treat the Mulan Assessment as another language: MBTI describes preferences; the Mulan Assessment observes choices in specific situations. Neither replaces the other, and you do not have to pick one.
20+ registered trademarks · software copyrights · 30+ novel copyrights · proprietary engine modules—building a multi-dimensional IP moat.
20+ fully registered graphic trademarks, including the primary brand logo and auxiliary logos, covering all classes and multiple industries—our identifiable core visual assets.
10+ registered word trademarks, headlined by "MulanXu®" and "Mulan Ci," covering key classes including 9/35/41/42—complete brand protection.
software copyrights covering core technology modules such as the AI emotional engine, multi-character interactive systems and content generation platforms.
30+ original novel copyrights, spanning the MulanXu® novel series and derivative works—the content source of the AI emotional interactive drama matrix.
MulanXu® takes emotional interactive content as its core strategic direction: long-form novel narrative assets + self-built emotional engine + offline scenario assets — channeled into one storyline, turning users from "story watchers" into "story rewriters."
A platform company in the emotional interactive content sector — long-form IP pipeline + multi-drama matrix architecture. From Mulan Xu to Mulan Dian to Mulan Tea Story, one unified tech stack powers the entire track.
Every character is driven by an independent AI model. Seven AI companions each have their own personality, memories, and growth arcs; 151 decision points generate plot branches in real time, and five endings are written by user decisions.
One IP, three monetizations. Long-form novels accumulate narrative assets, AI emotional interactive content carries online Token payments, and offline club/venue scenarios complete the landing—content invested once, reused continuously across multiple ends.
From one drama to an ecosystem. Themes expand from historical drama to urban and sci-fi; AI characters grow from seven to many more; IP licensing and user co-creation open gradually. The Emotional Interactive Engine is now open to creators and enterprises.
Track A · Organic Growth + Track B · Strategic M&A · Locking in 16 million users by 2027
Pre-A Round in Progress · Investment & media: pr@mulanxu.net
HR email: hr@mulanxu.net
📍 Canal Vanke Center, Gongshu District, Hangzhou, Zhejiang
MulanXu® HQ · next to Canal Central Park · direct Metro Line 5 to Gongchen Bridge East Station
The MulanXu® headquarters building awaits your visit.
A brand-new species born of the strategic fusion
of two seasoned teams, each twenty years deep in its field.
The MulanXu headquarters building awaits your visit.
Built on deep foundations, rising with full force—
leading the new track of emotional technology. — MulanXu · Corporate Vision
