Selected work

Role

Built recommendation engines, dynamic price propensity models, monitoring, retraining workflows, and analytics dashboards.

Environment

Commercial systems where customer targeting, personalization, and pricing decisions had to connect directly to business outcomes.

Result

• 10% revenue increase
• Stronger personalization and targeting
• Better visibility through monitoring and dashboards

Constraint

The models had to fit live business decisions, not sit beside them.

Role

Led deployment of recommendation, churn, and demand-forecasting models with CI/CD, MLflow, Kubernetes, and stronger MLOps standards.

Environment

Regional markets across North America, Europe, and Asia, affecting roughly 20 million customers.

Result

• 15+ models deployed
• Cross-region rollout reaching roughly 20 million customers
• 50% faster deployment timelines from better MLOps discipline

Constraint

Consistency and scale mattered as much as the underlying model performance.

Role

Built transformer-based speech-to-text and speaker-identification systems integrated into live operational workflows.

Environment

Real-time call-center analytics where latency, production integration, and operating cost all mattered.

Result

• 98% transcription accuracy
• 80% reduction in compute cost
• Stronger operational use of live language data

Constraint

Real-time systems make reliability and latency design decisions non-negotiable.

Role

Built an analytics system combining outage, GIS, weather, and operational data through spatiotemporal methods.

Environment

Distributed utility infrastructure where locating outages faster improves response speed and continuity of service.

Result

• 78% improvement in outage-location accuracy
• Detection time reduced from hours to minutes
• Stronger operational response

Constraint

This had to work inside mission-critical operations, not as a lab exercise.

Role

Built an end-to-end inspection system using drone imagery, synthetic data generation, and computer vision models for defect detection.

Environment

Infrastructure inspection with limited labeled defect data, high manual review cost, and real field-safety implications.

Result

• 92% defect-detection accuracy
• 70% reduction in manual inspection time
• About $3M in annual operational savings
• Safer inspection workflows

Constraint

The system had to overcome the data bottleneck without lowering reliability.

Role

Built retrieval and generative AI systems for technical and compliance-heavy documentation in regulated operating teams.

Environment

Teams dealing with dense standards, operating procedures, and knowledge that had to be usable at the point of work.

Result

• Faster information retrieval in operational use
• Stronger compliance support and response quality
• Up to $2.8M in projected annual savings

Constraint

The system had to make knowledge usable without adding friction or eroding trust.

Role

Built model deployment systems, HPC workflows, geospatial analytics pipelines, and cloud architecture patterns for scale and reliability.

Environment

Large analytics and training workloads, including 16TB/day geospatial pipelines and GPU-heavy AI infrastructure on Azure and AWS.

Result

• 90% reduction in compute cost
• 35% cloud cost reduction on geospatial analytics platforms
• Better path from experimentation to deployment

Constraint

In serious systems, infrastructure is not support work. It is part of the product.