Evidence
Separate prediction from proof. Define what is known and which experiment must come next.
AI DRUG DISCOVERY / INDUSTRY RESEARCH / VENTURE BUILDING
From scientific
promise to
executable systems.
I study how AI moves beyond impressive outputs and becomes a verifiable operating capability across discovery, development, clinical translation, and regulatory value.
AN HUIResearcher · Operator · Founder
0139 original essays19 company studies20+ years in pharma & ventures
01 — WORKING THESIS
The question is not whether AI can generate a molecule.
The question is whether each critical decision can be tested earlier.
Drug discovery is an evidence architecture. A binder is not yet a drug. A target ranking is not causal proof. Durable advantage appears when models are embedded in a system that learns from experiments, manufacturing, clinical outcomes, and failure.
02 — HOW I READ THE FIELD
Four lenses.
One operating discipline.
Frontier technology becomes valuable only when evidence, workflow, ownership, and incentives align.
System
Connect data, models, wet labs, automation, CMC, and clinical learning.
Execution
Translate constraints into priorities, teams, budgets, partnerships, and measurable gates.
Patient
Judge progress by better decisions, higher probability of success, and patient value.
03 — RESEARCH SIGNALS
Following the points where AI changes the probability of success.
Target intelligence
Genetics, phenomics, multimodal data and causal evidence
↗Molecule design
Structure, proteins, antibodies, small molecules and delivery
↗Closed-loop R&D
DMTA, robotics, physics engines and failure-data feedback
↗Clinical translation
Biomarkers, pathology, MRD, trial design and patient matching
↗Platform economics
Software, partnerships, milestones, assets and regulatory proof
↗
VENTURE PRACTICE / 1999—NOW04 — OPERATOR’S PERSPECTIVE
Built in the real world,
not only observed.
My perspective is shaped by pharmaceutical commercialization, a multinational pharma environment, laboratory construction, and serial entrepreneurship.
At Annray Testing, I helped build a professional laboratory from zero. At Yundi Technology, I work on strategic positioning, product direction, value proposition, and the transition toward AI-native products.
1999Pharmaceutical market practice
2002Roche China · business & sales
2005Co-founded Meiren Technology · first venture
2011Co-founded Annray Testing · built a ~4,000㎡ lab in 3 months, led CMA/CNAS accreditation, scaled to ~120 people and ¥30M annual revenue in 4 years
2019Co-founded Yundi Technology
2025New Leaf AI Drug research
05 — FIELD NOTES
Selected research
Original analysis published through New Leaf AI Drug. Articles open in WeChat and are currently written in Chinese.
Yundi Technology
Co-founder · Strategy, product direction, value proposition, and operations for the company’s AI-native transition.
AgentShow.pro ↗Annray Testing
Co-founder & General Manager · Built a ~4,000㎡ professional laboratory in 3 months, led CMA/CNAS accreditation and quality systems, scaled to ~120 people and ¥30M annual revenue in 4 years, and completed a successful equity transfer.
Annray Testing ↗“The real promise of AI drug discovery is not a system that always sounds right. It is a system that finds errors earlier and learns faster.”
Venture finance & capital practice
Finance lead across three ventures — Meiren Technology, Annray Testing, and Yundi Technology. Led the equity transfer negotiation and successful exit of Annray Testing. Holds Intermediate Accountant qualification.
Financial judgment for strategic decisions
Reads operating risk and opportunity through numbers, embedding budget discipline, financing rhythm, and exit planning into the company’s strategic choices.
“Numbers are not only a record of what happened. They are a lens for what is possible.”