Artefact · Telecommunications, insurance, pharmaceuticals
From models to systems other teams could operate
Data and ML consultant
- Largest portfolio
- 3M members
Three client missions moved from analysis or modelling into software used by sales, marketing, and data teams. Continuous internal engineering work ran alongside them and marked the start of my platform focus.
Orange: validating an agent with its sales users
A commercial agent explored a household's broadband and connectivity needs before proposing a personalised offer. I developed it with LangGraph on Vertex AI and instrumented it with MLflow and LangFuse.
The delivery included a validation phase with the sales teams who would use it. Their feedback supported acceptance and continuous improvement before wider use. The same mission also delivered a pipeline forecasting how many dwellings would become connectable to fibre across several time horizons and business-rule constraints.
Matmut: making a three-million-member segmentation operable
An end-to-end clustering pipeline segmented three million members by product holdings, loyalty, and development potential. Marketing teams could then design strategies around customer profiles rather than isolated campaigns.
Delivery covered the SQL, critical-function tests, documentation, and preparation for industrialisation. I also developed an application for inspecting the resulting clusters against the source data on HDFS.
Bayer: combining recommendation signals with business rules
A hybrid recommender combined collaborative filtering with business rules for commercial agents. The rules preserved requirements that collaborative signals could not override. The recommender reached production and was delivered to the sales teams.
Scheduled Airflow pipelines refreshed the data daily and fed the sales dashboards. The engagement also included coaching the client's data science teams on MLOps practices so they could operate the delivery after handover.
Internal engineering: the start of the platform work
The internal squad explored new engineering approaches and improved the firm's shared practices. My contributions included RAG evaluation tooling, a Terraform deployment path for applications behind Google Cloud IAP, an image-similarity proof of concept for eBay, and a purchase-propensity scoring pipeline.
Google Cloud belongs to these dated projects: the Orange agent ran on Vertex AI, and the deployment tool targeted IAP with a load balancer and custom domain. My current platform work is centered on AWS and Kubernetes.
Stack
- LangGraph
- LangFuse
- Vertex AI
- Gemini Pro
- MLflow
- LangChain
- OpenAI
- Airflow
- PySpark
- AWS Batch
- Terraform
- GCP IAP
- HDFS
- SQL