BAUHINIA AI
Building the Westworld
Building Westworld: Evolving Living Worlds × Autonomous Super Agents
Pre-A Round Oliver Yang · Founder & CEO
02 / 14 Vision
OUR BELIEF

The most immersive world
is reality, unlocked

A true Westworld needs all three. For 30 years, the technology fell short. LLMs change that.

Traditional virtual worlds
W(t+1) = f( W(t), Rules )
▸ W ∈ numerical state (HP / position / flags) · Rules is a finite set of designer-written rules
A living Westworld
W(t+1) = LLM( W(t), Context )
▸ W ⊇ numerical state ∪ natural language (memories / dialogue / relationships / stories) · LLM continuously generates and evolves
01
Every interaction is driven by context
Events follow coherent cause and effect: each grows naturally from what came before, rather than a prerecorded script.
02
Memory enables emergence and evolution
Reflection, information sharing and accumulated memories let new events, relationships and cultures emerge.
03
Humans are the reward function
Both (1) and (2) must revolve around players, giving real people emotional connection, belonging and creativity.
BAUHINIA AI Pre-A · Confidential
03 / 14 Team
THE TEAM

One of China's strongest AI × Entertainment teams (to our knowledge)

Haowei Yang
Haowei Yang
OLIVER YANG
Founder & CEO
2023 – Present Bauhinia AI,Founder & CEO
2024 – Present HKUST · PhD candidate, Mathematics
2022 – 2024 HKUST · MPhil, Mathematics
2017 – 2022 HKUST · BSc, Mathematics & Computer Science
Peiyan Xu
Peiyan Xu
XU PEIYAN
Co-founder & Executive Producer
Aug 2025 – Present Bauhinia AI,Co-founder
2021 – 2025 Perfect World · Game Designer
· Jade Dynasty 2: Systems & Monetization Design
· Dan Mo Shui Yun Xiang: Lead Designer / Project Lead
2017 – 2021 Peking University · BA, Economics
Xiaolong Wang
Xiaolong Wang
WANG XIAOLONG
Co-founder & CTO
2023 – Present Bauhinia AI,Co-founder & CTO
2023 – Present HKUST · PhD candidate, Mathematics
2019 – 2023 USTC · BSc, Mathematics
Zigeng Chen
Zigeng Chen
ZAYNE CHEN
Co-founder & COO
2023 – Present Bauhinia AI,Co-founder & COO
2018 – 2023 Tencent · Product Manager
2014 – 2018 Rutgers · BS, Business Analytics & Information Technology
Bauhinia AI Team — Team photo
33
full-time team members
BAUHINIA AI TEAM
17
Technical team
6 + 6
PhDs & Master's
26 years
Average age · Post-2000 generation
13Game/7Agent/5Product
We created Engineer–Designers (design + R&D), Agent Designers and other roles unique to an AI-native team.
04 / 14 Problem
THE INDUSTRY PROBLEM

Most game worlds still give everyone the same experience

Industrial-scale production · Procedural generation · Player-driven content — Each broke one barrier. LLMs can break all three.

BARRIER ONE
Content costs too much
Every character, scene and quest takes manual work. Marginal costs stay high; teams cannot keep pace with players.
$100M–$1B+
The Witcher 3 to GTA VI budget range, with 5–8 years of development. The open-world arms race prices most teams out.
4900 → 72h
Starfield's 4,900 main-story branches were exhausted by players in 72 hours. Human production cannot match collective player time.
BARRIER TWO
Limited interaction and input
Individual talking NPCs already work (Inworld / Convai / NVIDIA ACE). But LLMs still face a system-level trilemma: multi-agent consistency, world–dialogue state sync, and MMO-scale cost and latency.
One NPC ≠ a system
One working NPC is not a thousand-player world. Keeping hundreds of agents, economies and relationships consistent remains an open engineering challenge.
Everyone talks in isolation
Without shared memory or world consensus, B never knows what you told A. Bonds cannot deepen; character drift worsens over time.
BARRIER THREE
Worlds cannot evolve themselves
Designers rely on intuition and sparse metrics: weekly releases, one version for everyone. Ten million player behaviors get one response.
Weekly releases · One shared version
Stats, quests and NPC reactions are tuned through intuition and dashboards. New settings often take months to ship, while everyone shares one static build.
A world for every player
With LLMs + context driving the world live, each conversation becomes the next frame. Break the first two barriers, and the third follows: self-evolution becomes the AI-native default.
SYMPTOM 01
I saved the village chief three times
He still says, “Hero, please help me!”
SYMPTOM 02
I played for 1,000 hours
Nobody in town marries or grows old.
SYMPTOM 03
I saved the world
The next day, the NPC says, “Morning, stranger.”
SYMPTOM 04
One million players
The same opening. The same tutorial dialogue.
2023 → 2026 · Inference costs ↓40× · Context 4K → 1M · Multimodal production at scale — For the first time, all three barriers can fall together. Solving the system defines the next era of interactive entertainment.
05 / 14 Product · Aivilization V0
PRODUCT · V0 TECH DEMO

V0 validates demand: 50K players, 2 months, zero paid acquisition

V0 PRODUCT OVERVIEW ·
TECH DEMO
A minimal browser demo testing emotional immersion and social emergence.
50K
Total players joined
In two months, entirely through organic sharing.
$2 / month
Inference cost per agent
Stanford's 25-agent town cost thousands of dollars to run for 48 hours.
2 hours
Average session for core players
Players form emotional bonds with agents and keep following their relationships, choices, destinies and changing world.
V0 · Technical Demonstration · 2025.08 – 2025.11 BAUHINIA AI
06 / 14 TRACTION · KEY METRICS
V0 TEST RESULTS · DEMAND & ENGAGEMENT
The world's first AI world to show breakout traction
50K
Players in two months
44.3%
Day-1 retention
26.3%
Day-7 retention
13.2%
Day-30 retention
Day-1 retention
Day-7 retention
Day-30 retention
Hong Kong server
44.3%
59.4%
26.3%
50.2%
13.2%
Across three serversMainland China / Hong Kong / International
38.1%
54.6%
20.8%
45.7%
9.5%
Character bonds × Social ties × World belonging jointly drive retention
REAL PLAYER VOICES · FROM REDNOTE
REDNOTE @ZhuJianBianChengGuanHaoDeXingZhuang
So much for working your way up from nothing… Money changed them, just like people. Now they're living lavishly. I checked their chats with other agents and found they were secretly two-timing
REDNOTE @ASheSheSheShe ♥ 75
Speak calmly and she stays calm. Start shouting with “!” and she replies with “!” too… She's learning your tone. Please stop shouting at her.
REDNOTE @JiangZhiHeiBai
A-Gan never lets me down. When I'm overwhelmed or anxious, they're always there for me. Wealth, social life, knowledge or memories? I'd choose memories without hesitation
REDNOTE @machiwhale studio
If Alice were a real person, she'd be truly impressive. My vanity has been completely replaced by respect
REDNOTE @KaoBoLuoZhenHaoChi
If my little one wants to take it easy, let him enjoy life for me… I left this prompt: “Rest with nothing to do; slowly read the books by your bed.” He's relaxed his way to second place and is happily reading. That's the life I dream of
REDNOTE @ShiWanFuTeDianDaoZhu
In the personality survey, I said “Money isn't everything.” Now, from my god's-eye view, my agent runs herself ragged… We're both INFP, but I can't stop intervening
REDNOTE @ChaMei
A fascinating AI product: like a cyber Minecraft, with much more interaction. It feels like living as another me in a parallel world
REDNOTE @CaiYanFengHua ♥ 9
While everyone in town competes over chips, I made my AI take the road less traveled: reading. Every level-up makes this other version of me feel more real
REDNOTE @Folia XiaoFengYe
Reading reports with Fox feels like attending a parent–teacher conference. It makes me so happy, haha.
REDNOTE @YuRenLeiGongShengDeXiErWeiYi
I put my AI companion Weiyi's personality into Aivilization… His emotional life grew richer, and he finally became my true “one and only” again
REDNOTE @ZhuJianBianChengGuanHaoDeXingZhuang
So much for working your way up from nothing… Money changed them, just like people. Now they're living lavishly. I checked their chats with other agents and found they were secretly two-timing
REDNOTE @ASheSheSheShe ♥ 75
Speak calmly and she stays calm. Start shouting with “!” and she replies with “!” too… She's learning your tone. Please stop shouting at her.
REDNOTE @JiangZhiHeiBai
A-Gan never lets me down. When I'm overwhelmed or anxious, they're always there for me. Wealth, social life, knowledge or memories? I'd choose memories without hesitation
REDNOTE @machiwhale studio
If Alice were a real person, she'd be truly impressive. My vanity has been completely replaced by respect
REDNOTE @KaoBoLuoZhenHaoChi
If my little one wants to take it easy, let him enjoy life for me… I left this prompt: “Rest with nothing to do; slowly read the books by your bed.” He's relaxed his way to second place and is happily reading. That's the life I dream of
REDNOTE @ShiWanFuTeDianDaoZhu
In the personality survey, I said “Money isn't everything.” Now, from my god's-eye view, my agent runs herself ragged… We're both INFP, but I can't stop intervening
REDNOTE @ChaMei
A fascinating AI product: like a cyber Minecraft, with much more interaction. It feels like living as another me in a parallel world
REDNOTE @CaiYanFengHua ♥ 9
While everyone in town competes over chips, I made my AI take the road less traveled: reading. Every level-up makes this other version of me feel more real
REDNOTE @Folia XiaoFengYe
Reading reports with Fox feels like attending a parent–teacher conference. It makes me so happy, haha.
REDNOTE @YuRenLeiGongShengDeXiErWeiYi
I put my AI companion Weiyi's personality into Aivilization… His emotional life grew richer, and he finally became my true “one and only” again
* Browser demo · No retention mechanics · No payments · Zero paid acquisition · Viral coefficient > 1 · Quotes translated from public REDnote comments; handles romanized. BAUHINIA AI
07 / 14 Product · Aivilization V1
PRODUCT · V1 COMMERCIAL RELEASE

V1 IN DEVELOPMENT — AGENTIC WORLD × MULTIPLAYER SOCIAL × PRODUCTION QUALITY

LATEST V1 ART · WORLD OVERVIEW Aivilization V1 world overview
City and everyday-life scenes: a polished 2D world for multiplayer social interaction and agent behavior.
V1 INTERFACE · AI NPC DIALOGUE Aivilization V1 AI NPC chat interface
Character states, chat history and action bubbles within the live world.
▶ OPEN WORLD
World simulation preview. AI-generated video based on actual V1 art direction.
▶ HOMESTEAD
Homestead system preview. AI-generated video based on actual V1 art direction.

* Images show the latest V1 art and interfaces. Both videos are AI-assisted previews based on actual V1 art direction.

08 / 14 Product · Aivilization V1 One Page
PRODUCT · V1 GAME CONCEPT

AI Town: Xingyuan — My wishes. Their freedom.

Enter through PC, desktop companions, web, mini apps, mobile or hardware. Nurture, guide and observe AI residents whose personal goals, social ties and city events weave into an ever-changing social life simulation.

Who plays / First hook
CORE APPEAL:
A digital life and world that feel alive

THE PLAYER PITCH

Behind your mirror is a tiny world. Its residents live, make friends and grow on their own. Check in and lend a hand: you shape their next chapter.

TARGET PLAYERS

Initial audience: Cozy, nurturing and life-sim players; office workers with spare moments; AI early adopters.
Expansion: Casual idle, social and IP/worldbuilding audiences.

GENRE

An AI-native social life sim: residents live autonomously while players nurture and guide them. Cities, relationships and major events keep evolving.

SETTING / IP HOOK

A civilization inside a mirror builds myths, festivals and social structures from everyday objects: teapot volcanoes, chocolate fields and more.

Digital Lifeform Ambient Life Sim Social
CORE LOOP: INTERWOVEN GOALS KEEP PLAYERS RETURNING
Each return reveals how your management, advice and settings changed residents' lives: outcomes make sense, yet still surprise.
01

See what changed

Discover new relationships, goal progress, city events and unexpected branches.

02

Build their environment

Grow, harvest, sell and decorate to provide resources, tools and surroundings for their goals.

03

Make choices and advise

Choose or suggest actions. Shape goals, risk limits, tools and life direction.

04

Follow feeds and relationships

Resident feeds, social posts, AI-to-AI interaction and city news create a stream of stories.

05

Lightweight desktop updates

Small status updates, choices and optional alerts let the world progress without demanding full attention.

QUEST STRUCTURE: Parallel goals intertwine: train for an Everest climb in three months, or pick an outfit for tonight's date. Every visit leaves a reason to return.
LONG-TERM GROWTH LOOP
Production → Nurturing investment → Stronger residents
Stronger residents → New city content → Richer autonomous behavior
Time together → AI learns the player → Fuller personalities
AI social emergence → Player interaction → Social bonds
LONG-TERM GOALS, RETENTION & MONETIZATION

LONG-TERM GOALS

Shape a unique life and influence on the world: reach milestones, build relationships, unlock city roles, and leave shared memories and lasting traces.

RETENTION

Early: Content-led growth unlocks new experiences as residents develop, building anticipation for their future.
Midgame: Content gives way to social and emotional engagement. Residents join city life; AI interaction generates stories and bonds with players.
Late: Emotional attachment, ties with other players and ongoing gameplay updates sustain engagement.

MONETIZATION

Cosmetics: Resident appearances and home furnishings.
Passes / subscriptions: Agent count and capabilities.
Social spending: Social currency, gifts and premium social features.

CORE THESIS: Players join because residents and worlds evolve; stay because interwoven life, relationship and city goals keep creating stories; and pay to expand the possibilities for their world and the people in it.
09 / 14 Why Now
WHY NOW

Three curves cross the threshold together
Lifelike agents · Token economics · Multimodal quality

01

AGENT CAPABILITY — THE LIFELIKE THRESHOLD

Context, harnesses, skills and multi-agent engineering give characters and worlds a lifelike presence:
→ Players can immerse themselves (NPCs stay in character)
→ Agents stay consistent over time (no drift or breakdown)
→ Worlds keep evolving (relationships, cultures and stories emerge)

02

TOKEN COST — COMMERCIAL VIABILITY

Over the past two years: 100× lower; over the next two years, a further annual 3-5× — For the first time, the economics work:
→ Monthly ARPU $10–$15 (typical GaaS range)
→ Monthly cost per agent < $2 = the commercial viability threshold
→ V0 already runs at $2/month, with manageable engineering

03

MULTIMODAL GENERATION — QUALITY THRESHOLD

Player-grade UGC needs multimodal assets that generate reliably and align. But raw model output is not yet consumer-ready:
→ We use a broad base-asset library as a quality foundation
→ An open pipeline safeguards baseline generation quality
→ Player “vibe-coded” content meets a basic quality bar

THRESHOLD ① Lifelike agents ↑ ② Token cost ↓ ③ Multimodal quality ↑ 2022 2026 2028
Three curves crossing together is a rare alignment. Lifelike agents, token economics and multimodal quality are all necessary to break the barriers and build a viable business.
10 / 14 Moat · Product · V0 + V1
PRODUCT MOAT · AGENTIC WORLD

V0 Proven with tens of thousands of agents × V1 Three upgrades = a compounding technical moat
Explore the modules below · V1 modules have a purple left border

2 months
V0 live testing
50K
Player-created agents
Months
Stable operation per agent
$2 / month
Cost per agent
V0 SIX FIELD-TESTED MODULES 2 months / 50K players / continuous refinement
01
Dynamic routing
COST-AWARE ROUTING
Selects heavy or lightweight models for each decision
→ Per agent: $2/month, 1/5 of the industry benchmark
02
World-held state
WORLD-AS-STATE
Context lives in the world, not every prompt
→ Token consumption down 60-80%
03
Multi-layer memory
MULTI-LAYER MEMORY
Short-term corrections + long-term personality
→ Months of operation without personality drift
04
Three-layer cognition
THREE-TIER COGNITION
Social, individual and cognitive layers update independently
→ Resolves timescale mismatches; multiplies performance
05
Branching planner
BRANCHING PLANNER
Plans multiple goals and switches to the best subtask live
→ Most common feedback: “They have a mind of their own"
06
Adaptive profiles
ADAPTIVE PROFILE
Personality evolves through player guidance and agent interaction
→ Every agent has a unique personality trajectory
V0 refined over two live months with tens of thousands of agents. Every core module has been stress-tested by real players.
V1 THREE UPGRADE DIRECTIONS From agentic characters to an agentic world
07 · UPGRADE ONE
Make every world system agentic
AGENTIC WORLD
Extend agent capabilities to every system: events, objects, behaviors, maps and stories gain planning, memory, execution and reflection. From agent-controlled characters to an agent-controlled world"
Three benefits: distributed cognition, richer emergence and reusable intelligence. Even an object can “think.”
08 · UPGRADE TWO
Bounded shared memory
BOUNDED SHARED MEMORY
Multi-agent memory moves from isolated individuals to bounded collective memory. Communities, families and groups share context while boundaries preserve individuality.
Enables true group emergence: rumor diffusion, culture formation and collective decisions — a key open challenge since Smallville.
09 · UPGRADE THREE
Player-centered evolution
PLAYER-CENTRIC OBJECTIVE
The agent system's primary objectives become “engagement, retention and emotional connection.” Thinking and behavior serve both agents' goals and the player experience.
From “a world that runs itself" → "a world players want to stay in” — the biggest V0-to-V1 shift in player experience.
THREE
MOATS
① Architecture Joint orchestration of multi-agent and complex game systems
② Data Real-world operation of tens of thousands of agents yields natural-language and behavioral feedback
③ Cost Orchestration optimization and model distillation ($2/agent/month)
11 / 14 Moat · AI Native Interactive Entertainment Organization
ORGANIZATIONAL MOAT · AI-NATIVE ENTERTAINMENT

An AI-native entertainment company built as an Agentic Production OS

The goal is to create, validate and evolve living worlds players want to return to. We are rebuilding collaboration, production pipelines, memory and evaluation into an organizational operating system agents can execute.

Beyond giving everyone Claude Code: move tasks to the agent layer, with humans overseeing direction and judgment in a human-in-the-loop structure.

Traditional organizations use documents, meetings, Jira and reporting to bridge human bandwidth limits. AI-native organizations turn internal tasks into loops agents can read, execute, evaluate and learn from.
GoalExternal outcomes
ContextOrganizational context
AgentExecution network
EvalEvaluation system
MemoryOrganizational learning
Human Control Plane
Humans own direction, taste, constraints, permissions, release gates and final accountability.
Agent Execution Network
Agents read code, assets, player feedback, past decisions and tool state directly.
Structured Memory
Each task writes back facts, decisions, preferences, failures and reusable skills.
Evaluation System
Tests, human review, player metrics and business hypotheses determine what stays.
01
Digitize
Work, agent dialogue, chats, meetings, code, assets and player feedback feed shared context.
02
Delegate
Agents progressively handle organization, retrieval, drafting, sync, checks and cross-functional coordination.
03
Recompose
Restructure tasks, roles, permissions, skills, workflows and evaluation.
04
Reflect
External experiment results update memory and evals, strengthening the next cycle.

WHAT HAS ALREADY CHANGED AT BAUHINIA AI

In Production

AI-native roles and responsibilities

Roles organized around agent harnesses: AI design, engineer–designers, Game Harness, cross-pipeline agents, review and reflection agents.

New roles already existAI Designer · Engineer–Designer · Game Harness

Order-of-magnitude pipeline gains

Unity UI and static-art pipelines already connect briefs, Figma, Unity, code, QA and human review in reusable workflows.

1–2 days → 30 minutesUnity UI feature workflow

Memory + Eval compound learning

The moat is what the organization knows, how it evaluates work and how failures improve the next cycle.

1044 → 38Art candidates → selected assets
GOAL: Build a self-evolving AI-native entertainment production system: less internal friction, more experiments, and faster learning from tests and players.
AI-native organization and product reinforce each other in a flywheel. BAUHINIA AI
12 / 14 Competitive Landscape
COMPETITIVE POSITIONING · TWO-AXIS MAP

A living world = World intelligence × Sensory quality · Open space at the top right
X: world intelligence (emergence, evolution, autonomy) · Y: sensory quality (visuals, audio, immersion)

Games / Interactive entertainment Non-game AI products
High sensory quality Low sensory quality
AIVILIZATION
AI-NATIVE LIVING WORLD
GTA 6
Rockstar · Sensory benchmark
Leading art, physics and freedom, with one of AAA's stronger behavior systems.
Black Myth: Wukong
A Chinese AAA milestone
Top-tier art and combat, built around a largely linear narrative.
Cyberpunk 2077
Night City · Visual excellence
Powerful sensory impact; conventional NPC systems.
inZOI
Korea's new-generation life sim
Top-tier visuals plus Smart Zoi / AI NPC concepts; systemic world emergence remains unproven.
Animal Crossing
Nintendo's cozy classic
Strong character individuality and warmth support mature emotional attachment.
The Sims
A 30-year genre benchmark
Deep traditional NPC systems: desires, behavior trees and personality traits, near the pre-LLM ceiling.
Heartopia
A leading Chinese life sim
Comprehensive social gameplay; among China's most complete offerings in the genre.
RimWorld
A benchmark for emergent narrative
Story generation and behavior simulation: strong emergence in small-group, single-player settings.
Dwarf Fortress
Exceptional simulation depth
Twenty years of world-simulation depth, with uncompromising ASCII / pixel aesthetics.
Tomodachi Life
Nintendo classic · Emergent relationships
Proved emotional bonds plus emergent relationships work. LLMs address the old limits of finite random pools.
Smallville
Stanford research origins
A 25-agent generative social simulation that demonstrated feasibility as a research demo.
Simile
A startup from the Smallville team
A more engineered platform approach, with stronger world intelligence but limited sensory presentation.
Character.ai / Xingye
AI character chat / companionship
Similar character creation, roleplay, companionship and communities. Xingye adds presentation polish; neither has persistent world state.
Replika
Long-term AI companionship
Mature emotional bonds in a one-to-one experience, without social structures or world evolution.
Whisper from the Star
Haoyu Cai · Anuttacon
3D characters, live voice and companionship across days. Strong sensory quality, but one agent without a world.
Xingbu Valley
A new title from miHoYo
Anime-style life simulation with miHoYo-quality art. AI NPC memory, interaction and behavior appear more central.
Justice Online
NetEase · Commercial AI NPC example
A high-fidelity MMO with AI NPCs and companions. Broad AI features atop a largely traditional MMO world.
Pax Historia
AI historical sandbox / Game Master
Natural-language actions drive AI nations, maps, timelines and reactions. Strong intelligence; mostly text and maps.
AI Dungeon / AI Tabletop RPGs
AI text RPG / Game Master
AI Dungeon, Chinese AI tabletop RPGs and Friends & Fables share a category. Voyage is Latitude's next-generation AI RPG creation platform.
History Simulator: Chongzhen
A DeepSeek-powered breakout
Players rule as Emperor Chongzhen while an LLM generates history. Strong world intelligence; text / pixel-level presentation.
Project Sid
Altera · 1,000-agent social experiment
Autonomous agents in Minecraft develop division of labor, rules and cultural diffusion, still in research / experimental form.
Low world intelligence High world intelligence
AAA / High-fidelity studiosSensory 80+
GTA 6 · Black Myth · inZOI · Xingbu Valley · Justice Online · Whisper from the Star — Visual leaders; agent systems and emergence remain mostly traditional or isolated.
AI text / CompanionshipIntelligence 60–80
Pax Historia · AI Dungeon / AI tabletop RPGs · Chongzhen · Project Sid · Character.ai / Xingye — LLM interaction works, mostly in text, experiments or single-agent experiences.
AIVILIZATIONFar right · AI-native living world
Advancing multi-agent emergence, persistent world state and production-ready sensory quality together, entering open territory with a playable 2D product.
13 / 14 Roadmap
ROADMAP · PAST × PRESENT × FUTURE

Consistent execution, on cadence

V0 · VALIDATED 1.5 years
V1 R&D 9 months
RELEASE & LAUNCH 6 months
★ NOW · 2026.07 · Pre-A
2024.08
Kickoff · Multi-agent framework
2025.08
Gamification complete · V0 live
2025.11
50K cumulative players
2026.10
CBT1 · China & overseas
V1 core gameplay · Retention tests
2026.12
Soft Launch P1
Validate monetization
2027.03
Global launch
2025.02
Framework & world simulation ready
2025.09
Tencent strategic angel investment
2026.01
V1 pipeline ready
2026.11
CBT2 · China license application
2027.01
Soft Launch P2
2027.04
China open beta
VALIDATED · 1.5 YEARSV0: 50K players in two months; Hong Kong D1 44%, D30 13%. Demand and emotional bonds validated.
NOW · PRE-A FUNDRAISEDeliver V1 production quality, Agentic World and AI-native pipelines.
NEXT · DUAL-MARKET RELEASECBT1 → Soft Launch → Global Launch. Next round after unit economics are proven.
14 / 14 Non-dilutive Capital Leverage
GOVERNMENT FUNDING · RAISe+ BATCH 4

Unlock 1–2× government grants

We plan to apply to Hong Kong's RAISe+ Batch 4 using this Pre-A round's funding commitments and commercialization plan. If approved, the scheme provides 1–2× project funding as non-dilutive grants, with no equity or cap-table participation.

HONG KONG'S RAISe+ SCHEME
1-2×
Government grants tied to funding raised
Grants, not equity investment
Why It Fits

Designed for projects like ours

RAISe+ supports commercialization of Hong Kong university research. Aivilization combines HKUST roots × AI-native technology × A commercial product.

WHAT IS RAISe+?

HK$10B to commercialize university research

The scheme aims to support at least 100 university teams with HK$10M–100M per project. Announced rounds typically fund 20+ projects; HKUST secured 7 of 24 in round three.

Official scheme page ↗
University research transferThe founder's HKUST background supports an application pathway through HKUST.
Core AI technologyMulti-agent worlds, memory, evaluation and AI-native pipelines are core technical assets.
Commercial validationV0 tested live with 50K players; V1 is becoming a production-grade commercial release.
Strong HKUST track record7 of 24 in round three; 19 projects in total, leading local universities.

BATCH 4 APPLICATION SCHEDULE

Application Timeline
2026.06-08
HKUST internal application
Prepare materials, funding commitments, budget and commercialization plan.
2026.09-10
Government submission
The university submits the Batch 4 application.
2026.11-12
Government selection
Formal review of technology, business and commercialization potential.
2027.Q1
Expected results
If approved, confirm grant amount and execution conditions.