Vision Language Models: Applications, Limitations, and Real-World Deployment
How VLMs are being applied across industries, from image search and captioning to robotics and visual question answering, and where the field is heading.
Artificial Intelligence Engineer · Shanghai
Building AI that works on the aircraft, not only in the lab.
Lucy Craddock is an AI engineer at the Airbus China Research & Development Centre, where she works on edge computer vision for cabin design and manufacturing. She has spent nine years taking machine learning from research into production for aerospace.
- Computer vision
- Vision-language models
- Surrogate modelling
- Explainable AI
- NLP
What she works on
Edge computer vision
Vision systems for cabin design and manufacturing innovation, built to run on edge hardware.
Vision-language models
Applying VLMs and large language models to industrial problems, and speaking publicly about where they work and where they fall short.
Aerodynamic surrogate models
Machine learning models that predict wing pressure distributions, fusing CFD and wind-tunnel data, with uncertainty quantification.
Explainable planning
Parallelised scheduling for constraint systems, built so that people can understand why a plan was chosen.
Experience
-
2024 – present
Artificial Intelligence Engineer
Airbus China Research & Development Centre · Shanghai, China
- Edge computer vision for cabin design and manufacturing innovation.
- Moved from Airbus Central Research & Technology in the UK in early 2024.
-
2021 – 2024
Artificial Intelligence Researcher
Airbus · United Kingdom
- Numerical analysis and surrogate models for aerodynamics, delivered into production.
- Data fusion of computational fluid dynamics and wind-tunnel data, with uncertainty quantification.
- Computed tomography for non-destructive testing.
- Explainable scheduling and planning, and NLP topic sentiment analysis.
-
2017 – 2021
Data Science Degree Apprentice
Airbus · Bristol, United Kingdom
- Designed and built a machine learning system that predicted wing pressure distributions better than simulations, and a variational autoencoder that cleaned the data automatically.
- Parallelised an explainable scheduling system for constraint systems using MARCO, Z3, PySpark and TensorFlow, deployed with Docker.
- Built the data pipeline and models for NLP topic sentiment analysis, and benchmarked deep learning against traditional methods.
Talks
Applying for a Degree Apprenticeship in Tech
A practical account of getting into data science through a degree apprenticeship instead of a traditional university route.
AI Ethics: Bias in Data
A lightning talk on how the data behind AI systems reflects the prejudices of the societies that produced it, with examples from crash testing, medical diagnosis and facial recognition.
Background
Education
- BSc Digital and Technology Solutions, First Class Honours University of Exeter, 2017 – 2021. Degree apprenticeship with Airbus.
- AWS Certified Cloud Practitioner Amazon Web Services, 2023
- Machine Learning Coursera, 2018
Tools and languages
- Python, Julia, PySpark TensorFlow, Keras, scikit-learn, pandas, NumPy
- Docker, gRPC, Linux, AWS Deployment from research prototype to production
- English and Mandarin English native; Mandarin limited working proficiency
Contact
The best way to reach Lucy is through LinkedIn.