Professional Experience
How the scope evolved
Not a second CV. This page shows how the work widened over time: from teaching and research, into consulting under constraints, and then into architecture, system ownership, and team leadership.
From research and teaching to full ownership of real-time ML systems
This page is intentionally not a second CV. It is the story of how my scope changed over time: from explaining ideas, to building models, to owning the systems, trade-offs, and teams needed to make those models useful in production.
- 🎓 Research roots
- 🏗️ Production systems
- ⚙️ Streaming architecture
- 👥 Team leadership
- 🏛️ Standards work
✨ What Changed Along The Way
- Early years: explaining and teaching I learned to break complex ideas down clearly and help others build confidence in technical subjects.
- Consulting years: delivery under ambiguity I learned how messy systems, unclear requirements, and real stakeholder pressure reshape “correct” engineering.
- Current years: systems and leverage I now focus on architecture, operational quality, and building teams that can ship dependable ML systems repeatedly.
teaching, research, communication
🎓 Teaching, doctoral work, and the habit of clarity
Before I was responsible for production systems, I spent years teaching and researching, which is where I developed the habit of explaining difficult ideas simply and structuring technical work carefully.
What I was doing
- Teaching Java, Python, MATLAB, HTML, CSS, SQL, AI, systems, and data structures.
- Completing doctoral research in persuasion dialogues, opponent modelling, and large knowledge graphs.
- Working close to formal methods, graph reasoning, and research-driven problem solving.
What stayed with me
- Technical communication is a force multiplier.
- Good systems thinking starts with clean abstractions.
- Explaining something clearly is often the best test of understanding it.
shipping under constraints
🏗️ Consulting became the bridge from data science to ML engineering
This was the period where “interesting model work” stopped being enough. I had to deal with enterprise constraints, legacy systems, production expectations, and the uncomfortable gap between experimentation and deployment.
Representative client work
- 🏦 UBS: graph analytics, process mining, and real-time insight pipelines with Kafka, Elasticsearch, and Python.
- 🚜 CNHi: time-series forecasting for agricultural vehicles with alerting and deployment paths.
- 📱 Vodafone: internal MLOps platform work on GCP and Kubeflow, with CI/CD, telemetry, and reproducibility.
What this phase taught me
- Most ML failures are systems failures, not modelling failures.
- Ambiguous environments are where architecture matters most.
- Bridging DS and engineering is a delivery problem as much as a technical one.
real-time systems at scale
🚢 Vortexa: owning streaming-first ML systems end to end
At Vortexa, the center of gravity shifted again: from delivering components to owning systems, quality bars, and the teams responsible for keeping them reliable over time.
What I built
- Destination and ETA systems using transformer-based sequence models with automated refresh and longitudinal evaluation.
- Real-time Kafka/Flink pipelines over global AIS feeds, powering prediction, anomaly triggers, and downstream decision support.
- AIS denoising operators based on Kalman filtering to improve signal quality and downstream model accuracy.
- MLOps foundations around rollout strategy, observability, model versioning, and operational discipline.
What I own now
- Cross-functional delivery across ML, DS, and DE.
- Hiring, mentoring, roadmap alignment, and execution quality.
- Architecture decisions that account for latency, replayability, and failure modes.
- Raising the bar from “the model works” to “the system can be trusted.”
Standards and research
- Since 01/2021 · ISO JTC 21 WG3
Committee Expert Member working on AI standards aligned with international and EU policy directions. - Since 10/2024 · UCL Department of Information Studies
Associate Researcher supporting AI application, standardisation, and ethics initiatives.
Talks and interviews
- 2023 · Agile in Action
Podcast interview on agile data science and the Vortexa journey. - 2022 · ODSC
Industry talk on dynamicio and abstracting I/O for ML systems. - 2020 · iunera & Big Data Warsaw
Interview and conference talk on agile data science and graph-driven analytics. - 2018 · Connected Data London & Minds Mastering Machines
Panel and talk appearances on graph AI and doing data science the agile way.
What has remained constant
- Production is the only truth.
- If it cannot be measured, it is not done.
- Deterministic systems beat clever hacks.
- Models must degrade gracefully.
- System quality should scale through people and process, not heroics.
Guiding principle: “Make it work. Make it right. Make it fast.”