From Experimentation to AI Adoption
IATA World Data Symposium
Moving customer-facing AI past the prototype stage, and the layers underneath it that most teams underestimate. Invited talk.
Shirley He · Zurich
Data scientist turned AI product lead.
I run the AI portfolio for customer experience at SWISS, and I publish the evals for everything I build.
- 7 years building models and pipelines
- Winner, Agentic AI Hackathon 2025
- Instructor, Constructor Academy
Selected work
Projects
Every project shows the evals and what failed.
01
Enterprise
Airline AI assistant
Growing an airline group's chat assistant from answering questions into orchestrating agents, with loyalty built in.
Read the case study →
02
Live · v0
ragpatterns
Seven RAG patterns, one bookshelf, nine questions with answers written before any pipeline ran. Most teams pick a pattern by accident.
Read the case study →
03
Live
Make My Gate
You land at D33 and fly out of A63 in 50 minutes. Will you make it? A Zurich connection planner: graph plus orchestrator scored 100% on 25 test questions, naive RAG scored 28%.
Read the case study →
04
Live
fixmybanana
Upload a handstand photo and a vision model calls it: good form or banana back. A joke app with a serious eval: 54 test photos, 60% agreement with my labels.
Read the case study →Recognition
Winner, Agentic AI Hackathon
Lufthansa Group Digital Hangar, with IBM, Microsoft and Cognigy, May 2025. Our team built RetentionAI: one agent finds the moments where personal attention changes the outcome, another shapes the gesture to fit the customer. Picked from 10 teams and 100 participants.
Speaking
Rooms I've been in.
Industry conferences, panels and internal sessions, in Europe and Asia. What worked, what didn't, and the numbers behind both.
Agentic AI in Practice
Lufthansa Group Digital Hangar, with Frankfurt Data Science and Netlight
Embedding autonomy and reasoning into AI systems. Co-presented with Christian-Heiko Ritter, then a hands-on campfire session that ran out of chairs.
Breaking Barriers and Redefining Leadership
Female Digital Pioneer · Ambition Pitch
The glass ceiling, what it actually takes for women to move into technology leadership, and the communities that make it possible.
Topics I take
- Agentic AI in production
- How to tell whether an AI product works
- AI in aviation
- Data science to AI product
- Women in technology leadership
- EN · DE · 中文
Also on offer: The January Problem, on why AI strategies fail the way New Year's resolutions do.
About
From coaching to code.
Before data science I was a personal trainer. I still train friends, and I measure success the way a coach does: by what someone can do now that they couldn't before.
Today I'm AI Product Lead at SWISS, part of Lufthansa Group's Digital Hangar. I run the AI portfolio for customer digital experience and decide which ideas get funded and which get stopped. Because I came from data science, I read every pitch from both sides: does it hold up technically, and will it survive real users, real data and real constraints?
Before that: an MSc in Data Science from Barcelona GSE, then seven years in the engine room building models, pipelines and cloud migrations. I work in English, German and Chinese.
Outside work I teach at Constructor Academy in Zurich and lead sponsorships for Women in Data Science Zurich. I have also taught HTML and Python to kids aged 7 to 17 with CoderDojo, which remains the hardest audience I have had to explain anything to.
Also on my desk
10 AI Workflows for Product Managers
Prompts, examples and quality checks for the work PMs do every week. Pre-orders open, delivered October 2026.
Contact
Have an AI product that needs to work?
I take on freelance AI consulting, workshops and speaking. Tell me which one you're here for.
shirleyishuman@gmail.com