‹
EN
OPEN
Astral Intelligence (A*I)
星际智能(A*I)
AI for Scientific Reasoning
面向科学推理的人工智能
Accelerating discovery by reducing human coordination bottlenecks
通过减少人类协作瓶颈加速发现
Founder: Peter Wang
创始人:Peter Wang
Prior exit: Sequents
此前退出:Sequents
Science is not compute-limited. It is reasoning-limited.
科学不是算力受限,而是推理受限。
Data abundance, insight scarcity
数据充足,洞见稀缺
Fragmented domains, siloed expertise
领域碎片化,专业知识孤岛化
High uncertainty, irreversible decisions
不确定性高,决策不可逆
Tooling optimized for analysis, not thinking
工具偏分析,不偏思考
The bottleneck is no longer information — it is alignment.
瓶颈已不在信息,而在一致性。
What actually slows discovery
真正拖慢发现的原因
Human coordination failure
人类协作失灵
Too many competing hypotheses
竞争假说过多
Weak synthesis across data and models
数据与模型之间整合薄弱
Slow belief updates
信念更新缓慢
Lost negative results
负结果丢失
Decisions justified after the fact
事后合理化决策
Why now
为什么是现在
Three shifts converged
三股变化汇合
Multimodal AI can reason over structure
多模态 AI 能在结构上推理
Language models can operationalize intent
语言模型能将意图转化为操作
Biology crossed a combinatorial threshold
生物学已跨越组合爆炸门槛
Scientific tools did not fail — human workflows did.
科学工具没有失败,失败的是人类流程。
Our insight
我们的洞见
AI should not replace scientists. AI should accelerate how scientists think together.
AI 不该取代科学家,而应加速科学家共同思考。
A*I focuses on
A*I 关注
Hypotheses
假说
Evidence
证据
Uncertainty
不确定性
Belief updates
信念更新
Not about
不是关于
Black-box predictions
黑箱预测
Autonomous discovery claims
自治发现的夸张宣称
Replacing scientific judgment
替代科学判断
What A*I is
A*I 是什么
A hypothesis-centric scientific reasoning engine
以假说为核心的科学推理引擎
Natural language questions
自然语言提问
Explicit hypothesis objects
显式的假说对象
Evidence aggregation
证据汇聚
Confidence tracking
置信度追踪
Full reasoning trace
完整推理链路
SQL, Python, and ML models live underneath, not on the surface.
SQL、Python 与机器学习模型在底层运行,而非界面表层。
What A*I is not
A*I 不是什么
Deliberate constraints
有意设定的边界
Not a wet-lab company
不是湿实验室公司
Not promising biological outcomes
不承诺具体生物学结果
Not replacing scientists
不替代科学家
Not selling black-box predictions
不售卖黑箱预测
Not a consumer or hardware product
不是消费级或硬件产品
Not replacing existing bioinformatics stacks
不替换现有生物信息学体系
Mental model
心智模型
From question to belief update
从问题到信念更新
Hypothesis made explicit
假说显式化
Operationalization into analyses
转化为可执行分析
Evidence gathered
证据收集
Confidence updated
置信度更新
Competing hypotheses compared
对比竞争假说
The unit of progress is belief convergence, not answers.
进步的单位是信念收敛,而非答案。
Core primitives
核心原语
Hypotheses (versioned, comparable)
假说(版本化,可比较)
Evidence (data, models, experiments)
证据(数据、模型、实验)
Belief states (confidence + uncertainty)
信念状态(置信度 + 不确定性)
Reasoning traces (auditability)
推理链路(可审计)
Sequents DNA, evolved for scientific decision making.
Sequents 基因,面向科学决策升级。
Architecture overview
架构概览
Hypothesis and evidence layer (A*I-owned)
假说与证据层(A*I 自有)
Tool execution layer (SQL, Python, bio tools)
工具执行层(SQL、Python、生物工具)
Model routing layer (frontier + open-weight)
模型路由层(前沿模型 + 开源权重)
Verification and audit layer
验证与审计层
Models are interchangeable. Reasoning is not.
模型可替换,推理不可替换。
Model strategy
模型策略
Frontier-powered, not foundation-dependent
依托前沿,而非依赖自研底座
Frontier models for raw reasoning
前沿模型用于高强度推理
Open-weight models for reliability, privacy, and cost control
开源权重用于可靠性、隐私与成本控制
Fine-tuning for discipline, not IQ
微调用于纪律性,而非智商
Bring-your-own-model for enterprise
企业可自带模型
Why biology first
为何先做生物学
The hardest domain
最难的领域
High uncertainty
不确定性高
Long timelines
周期长
Irreversible decisions
决策不可逆
Expensive mistakes
失误代价高昂
If A*I works here, it generalizes.
若 A*I 在此有效,即可泛化到其他领域。
Initial use cases
初始用例
Where reasoning breaks today
今天推理最易断裂之处
Aging and longevity biology
衰老与长寿生物学
Target validation in drug discovery
药物发现中的靶点验证
Cross-domain synthesis (genomics ↔ epigenetics)
跨领域综合(基因组学 ↔ 表观遗传学)
Internal scientific decision reviews
内部科学决策评审
Who the buyer is
谁是购买者
Enterprise and institutional users
企业与机构用户
Biotech startups
生物技术初创
Biopharma R&D
制药研发
Translational research labs
转化研究实验室
Computational biology groups
计算生物学团队
Not consumers. Not hobbyists.
不面向消费者,也不面向爱好者。
Market expansion trajectory
市场扩展路径
Same substrate, more domains
同一底座,扩展更多领域
Biology → materials science
生物学 → 材料科学
Biology → climate and systems modeling
生物学 → 气候与系统建模
Biology → engineering and policy
生物学 → 工程与政策
A*I is a reasoning platform, not a vertical tool.
A*I 是推理平台,而非垂直工具。
Business model
商业模式
Enterprise-first
企业优先
Paid pilots
付费试点
Institutional contracts
机构级合同
Expansion inside organizations
组织内扩张
Long-term platform licensing
长期平台授权
No freemium. No ads. No race to the bottom.
无免费模式,无广告,不打价格战。
Go-to-market
市场进入
Design-partner driven
设计合作伙伴驱动
2–4 serious design partners
2–4 家核心设计伙伴
Deep integration
深度集成
Measured reasoning acceleration
可量化推理加速
References before scale
先做口碑后规模
Trust precedes growth in science.
科学领域,信任先于增长。
18-month execution plan
18 个月执行计划
Four phases
四个阶段
Phase 0: framing and validation
阶段 0:框定与验证
Phase 1: core reasoning engine
阶段 1:核心推理引擎
Phase 2: design partner pilots
阶段 2:设计伙伴试点
Phase 3: institutional readiness
阶段 3:机构级就绪
Each phase de-risks the next.
每一阶段为下一阶段去风险。
Phase 0 (Months 0–2)
阶段 0(0–2 个月)
Foundation
基础建设
Product spec and primitives
产品规范与原语
Architecture finalized
架构定稿
Scientist interviews
科学家访谈
1+ design partner committed
至少 1 家设计伙伴承诺
Success: “This matches how we think.”
成功标准:“这符合我们的思考方式。”
Phase 1 (Months 3–6)
阶段 1(3–6 个月)
Private alpha
私有 alpha
Hypothesis formalization working
假说形式化可用
SQL and Python execution
SQL 与 Python 执行
Competing hypotheses supported
支持竞争假说
Inspectable reasoning traces
可检查的推理链路
Success: A real scientific decision influenced.
成功标准:影响一个真实科学决策。
Phase 2 (Months 7–12)
阶段 2(7–12 个月)
Validation
验证
2–4 active pilots
2–4 个活跃试点
Multi-user reasoning
多用户推理
Evidence persistence over time
证据长期沉淀
Measured convergence acceleration
可量化的收敛加速
Success: Behavior change without prompting.
成功标准:无需提醒即可改变行为。
Phase 3 (Months 13–18)
阶段 3(13–18 个月)
Institutional readiness
机构级就绪
Security and auditability
安全与可审计
Clear buyer identified
明确购买方
Paid pilot or LOI
付费试点或意向书
Repeatable adoption pattern
可复用的采用模式
Success: Ready for $6M follow-on.
成功标准:为 $6M 后续融资就绪。
Cost philosophy
成本哲学
Capital efficiency with credibility
在可信度下的资本效率
Senior, small team
资深小团队
No foundation model pretraining
不做底座模型预训练
Cloud-first, scale later
云优先,后扩展
Spend where trust is built
把钱花在信任建立处
18-month budget summary
18 个月预算概览
Category
Cost
Core team (6–7 people)
~$2.3M
ML compute
~$0.3M
Data and tooling
~$0.15M
Infra and DevOps
~$0.15M
Product and UX
~$0.2M
Legal and IP
~$0.15M
Pilots and partnerships
~$0.1M
Total: ~$3.3–$3.5M
总计:约 $3.3–$3.5M
Funding plan
融资计划
Staged conviction
分阶段建立信念
Initial: $500k–$3M to fund Phase 0 → Phase 2
初始:$500k–$3M,覆盖阶段 0 → 2
Follow-on: up to $6M triggered by institutional readiness
后续:最高 $6M,由机构级就绪触发
Investor optionality preserved.
保持投资人可选性。
Founder
创始人
Peter Wang
Peter Wang
Founder, Sequents (AI reasoning over structured data)
Sequents 创始人(结构化数据推理)
Full-stack and systems thinker
全栈与系统型思考者
Proven human-in-the-loop AI
有验证的人机协作 AI 经验
Comfortable at the intersection of AI, science, and institutions
擅长 AI、科学与机构交叉地带
Why A*I wins
A*I 为什么能赢
This is not a model race
这不是模型竞赛
Reasoning substrate is the moat
推理底座是护城河
Trust and auditability matter more than raw IQ
信任与可审计比原始智力更重要
Hypothesis-centric workflows are underbuilt
以假说为中心的流程仍很薄弱
Biology demands this approach first
生物学最需要此方法
The takeaway
核心结论
A*I is infrastructure for scientific thinking.
A*I 是科学思考的基础设施。
We do not sell answers
我们不卖答案
We sell better decisions
我们卖更好的决策
We accelerate convergence under uncertainty
我们在不确定性下加速收敛
If humans can think together faster, discovery follows.
若人类能更快共同思考,发现便会随之而来。
Appendix: optional deep dives
附录:可选深潜内容
Model routing policy (frontier vs open)
模型路由策略(前沿 vs 开源)
Example hypothesis flow (mock)
示例假说流程(模拟)
Security and privacy stance
安全与隐私立场
Long-term expansion scenarios
长期扩展场景
Regulatory fit and US environment
监管契合与美国环境