Evidence-first thinking
I start with the quality of evidence, study design, and explicit limitations before turning findings into claims.
Faculty of Medicine, University of Helsinki · Helsinki, Finland
Runhai He
I build medical AI systems that connect cell morphology, clinical variables, and survival outcomes.
I am a first-year doctoral researcher in medical AI at the University of Helsinki, working in Hematoscope Lab at Helsinki University Hospital and the University of Helsinki. My background combines applied mathematics, machine learning, computer vision, survival analysis, full-stack development, and scientific communication.
Recruiter snapshot
Fast fit signals
Target roles
Medical AI / Research Engineer / AI Product
Base
Helsinki, open to EU-focused opportunities
Proof
Research pipeline + shipped SaaS + client CMS
Links
CV, GitHub, LinkedIn, Scholar ready for review
1.66M
single-cell images analyzed
0.94
AUROC malignancy index
3.93
MSc GPA, rank 1/15
Medical AI pipeline
From cells to survival risk
Input
Encoder
Survival
Fusion
Input1.66M bone-marrow single-cell images + clinical variables
EncoderSelf-supervised ViT maps each cell into 384-dimensional embeddings
SurvivalMI-Net treats each patient as a bag of cells for risk prediction
FusionLate fusion with Cox models improves clinical-only C-index 0.73 -> 0.76
Modeling
ViT + MIL + Cox fusion
Scale
912 patients / 5,559 controls
Delivery
Next.js, Vercel, CMS systems
Research
First-author medical AI abstract
Profile
A technical researcher profile built for fast evaluation.
I am a first-year doctoral researcher in medical AI at the University of Helsinki, working in Hematoscope Lab at Helsinki University Hospital and the University of Helsinki. My background combines applied mathematics, machine learning, computer vision, survival analysis, full-stack development, and scientific communication.
Multimodal deep learning for cancer prognosis and medical image analysis
Survival analysis with Cox models, C-index, time-dependent AUROC, and risk stratification
End-to-end Python and PyTorch pipelines for large biomedical image datasets
Independent shipping experience with AI-powered SaaS and responsive web products
Structured execution
I break complex research and product problems into data, method, interface, and communication layers.
Reviewer-oriented output
I design artifacts so a busy reader can inspect context, methods, and outcomes without unnecessary friction.
Review path
A recruiter-friendly route through the portfolio.
The homepage now behaves like a review product: start with fit, inspect proof, then contact with context.
0130 sec
Scan fit
Use the snapshot, metrics, and role signals to decide whether the profile matches medical AI or AI product work.
Start at profile022 min
Inspect evidence
Open selected work to compare research pipelines, shipped products, implementation details, and measurable outcomes.
Review projects031 min
Check capability
Validate technical range across modeling, survival analysis, full-stack delivery, communication, and collaboration.
See skills04Direct
Contact with context
Use email, LinkedIn, GitHub, or Scholar links when the opportunity, team, and expected contribution are clear.
ContactExplore
Choose the section you need.
The portfolio is split into focused subpages so recruiters can jump directly to the relevant evidence.
Selected Work
Review portfolio projects, research tooling concepts, and technical presentation work.
Experience
Scan academic, technical, and evidence-oriented experience as a concise timeline.
Skills
See capabilities grouped by research, data, web implementation, and workflow.
Contact
Get email, LinkedIn, and GitHub contact paths without scrolling through the site.
Personal
Read a more human introduction with optional life photo, personal statement, and background details.