Drilling optimization
Predicting stuck pipes in real time to cut non-productive time on drilling rigs.
RoleLead Data Engineer, then Lead ML Engineer, then Lead Solutions Architect
Result~$37M a year in programme savings
The problem
When a drill pipe gets stuck, a rig stops: hours or days of non-productive time (NPT), at a very high cost.
What I did
A multi-year programme of several projects, where I was successively responsible for the data, the machine learning, then the solution architecture.
- Set the AI strategy for anticipating stuck pipes, with drilling experts, ML engineers and data scientists.
- Built real-time ingestion and prediction pipelines, deployed in the cloud.
- Integrated a partner’s NPT models (SLB) for some types of rigs.
- Added a language interface (RAG) so engineers could question the data in plain text.
- Shaped resources, planning, client alignment and technology costs (cloud, licences).
Result
The programme’s technology saved about $37M a year, reinvested in R&D to keep improving the AI tools.
Python · PyTorch · ClearML · RabbitMQ · Kubernetes · Terraform · Huawei Cloud · AWS · Azure · Angular · Keycloak · Llama