Victoria Eshelby · Data and AI
I fix the part that doesn’t add up.
Sometimes the data is wrong. Sometimes the process takes far longer than it should. Sometimes a system only works because one person knows how to keep it alive. Those are usually the problems I want to work on.
How I work
Four questions I keep asking
Job titles divide data work into neat boxes. Real problems rarely cooperate, so I tend to follow the question across those boundaries.
What does this data actually mean?
I work with the people asking the question before deciding the grain, dimensions, metrics, or model.
Where can it break?
I trace the data, test the assumptions, and make failure visible before somebody builds a decision on top of it.
Can somebody else use it?
Teaching changed what finished means to me. I build examples, checks, and documentation so the next person can use the work without needing me beside them.
What is AI actually useful for?
I use AI to speed up implementation, search, and prototyping, but I still test the output, check the cost, and make the decisions.
Selected work
The work I can talk about publicly
Production code and internal evidence stay private. These notes explain the problem, my part in it, and what I learned without pretending the whole result belonged to me.
Making player interaction data mean the same thing twice
I designed event models, dimensions, and navigation views for high volume telemetry, then built checks around the grain, mappings, identities, and journey logic most likely to go wrong.
- Reconciled source events with what reached the model
- Made unknown identities explicit instead of quietly dropping them
- Defined how sessions affect attribution
- Added regression checks for later changes
Finding out whether an AI idea was worth running
I built a Voice of the Customer proof of concept across internal and external feedback. It worked. The more useful result was finding that runtime cost, not technical feasibility, was the main barrier to scaling it.
Fabric · PySpark · Microsoft Foundry · Microsoft Copilot Studio
What 13 million learner records could tell us
I explored role and certification patterns across more than 13 million learner records and prototyped recommendations for people moving towards advanced role based learning.
Databricks · PySpark · Feature engineering
Learning a new industry quickly enough to be useful
At Metyis, I moved between forecasting, pricing, environmental risk, and sensor work across retail, industrial IoT, and aquaculture. The methods mattered, but learning the domain came first.
Python · R · pandas · PySpark
Trainer owned Azure labs for small workshops
Hosted lab platforms such as Skillable can be expensive or difficult to access for small initiatives. I could not secure internal funding, but I still wanted to support Maud's refugee learning initiative. I built a reusable Azure approach that trainers can run in a dedicated training environment or limited subscription. The first implementation supported 29 learner workspaces.
Azure · PowerShell · Entra ID · Microsoft Foundry · AI assisted engineering
Read the case studySpeaking and research
Things I have explained out loud
Training is not separate from my engineering work. It is where I learned to notice the assumptions other people cannot see yet.
Featured talk · YouTube
Intelligent Pipelines in Action: AI Collaboration with Fabric
A practical discussion of how intelligent workflows and Microsoft Fabric can support collaborative data and AI delivery.
Watch the talk on YouTubePresentation · YouTube
Microsoft Fabric Introduction: Your First Step Into the Unified Analytics Platform
An accessible introduction to Microsoft Fabric and its unified analytics capabilities.
Watch the presentationChat moderator · YouTube
How to Pass Exam DP 600: Implementing Analytics Solutions Using Microsoft Fabric
Supported the live learning experience as a chat moderator for this exam preparation session.
Watch the sessionPeer reviewed research
Stay Home and Stay Active? The impact of stay at home restrictions on physical activity routines in the UK during the COVID 19 pandemic
Coauthored research examining changes in physical activity routines during pandemic restrictions.
Experience
How I arrived here
2025 — Present
Analytics Engineer assignment
Microsoft · Mojang Studios collaboration
Designing governed telemetry models, reusable semantic assets, validation contracts, and implementation patterns for large scale game analytics.
2022 — Present
Senior Technical Trainer
Microsoft
Delivering advanced data and AI learning, creating technical demonstrations, mentoring professionals, and building proofs of concept across Azure, Databricks, Fabric, and AI platforms.
2020 — 2021
Data Scientist
Metyis
Building forecasting, pricing, data validation, and machine learning solutions across retail, industrial IoT, and environmental domains.
2020
Machine Learning Research Assistant
WMG, IDH Lab
Developing computer vision models and applying statistical analysis to public health and mobility research.
Credentials
The formal bits
Microsoft certifications
- DP 700 Fabric Data Engineer Associate
- DP 600 Fabric Analytics Engineer Associate
- PL 300 Power BI Data Analyst Associate
- AI 102 Azure AI Engineer Associate
- DP 100 Azure Data Scientist Associate
Education
- MSc Artificial Intelligence Utrecht University
- BSc (Hons) Psychology University of East Anglia
Contact
If something here made you think of a problem, tell me about it.
LinkedIn is the easiest place to reach me. You do not need a polished introduction. “I found the thing that does not add up” is enough.