Shilpi Bhargava

Oversaw development of geospatial intelligence AI workloads on telemetry data, including impound alerts and prediction models
trained on billions of records to deliver dwell insights; architected and built the geofencing service as the backbone of the platform.
Established CI/CD patterns for iterative delivery of AI/ML workloads, adopted by 3 teams, improving reliability and repeatability of
deployments across environments.
Introduced MLflow for experiment tracking and model lifecycle visibility, tracking model experiments to support consistent model
iteration and release management.
Built proof-of-concept workflows using LangChain to orchestrate LLM-based components for agentic AI and RAG-style interactions
aligned to platform use cases.
Enabled impound alerts and dwell insights to support operational decision-making, reducing time-to-action by 50%.
Technologies Used: LangChain, CI/CD, MLflow

Delivered a Python-based proof of concept for an Oil & Gas client applying deep reinforcement learning to real-time vessel routing
and scheduling, generating optimized schedules for 100+ tanker vessels and 200+ shipments while accounting for weather routing,
stochasticity, and dynamic decision-making; cut manual scheduling effort by 6 hours/week.
Applied MLflow to track deep reinforcement learning experiments and integrated CI/CD practices to streamline iteration and
reproducibility for the Python PoC, reducing rerun time by 20%.
Technologies Used: Python, Deep Reinforcement Learning, MLflow, CI/CD

End-to-end deployment of machine learning pipelines for models recommending strategies to optimize main engine fuel efficiency
and performance. Developed and deployed hull fouling prediction models for Maersk vessels using sensor data, with real-time
alert dashboards and API-based integration for shore and vessel applications.

Developed a Python Dash application to model warehouse outbound operations, enabling wave creation of case packs at
distribution centers to boost UDC productivity; successfully productionized after strong leadership reception. Mixed-Integer
Programming models for network planning to optimize store-to-DC allocations, minimizing transportation costs while accounting
for capacity constraints.
Technologies Used: Python Dash, Mixed-Integer Programming

Global Production Network Planning: Solved strategic production planning MILP in IBM ILOG CPLEX (OPL Studio) by conducting
Total Landed Cost evaluations to assess sourcing changes and global footprint impact. Built VBA-based Genetic Algorithm to
minimize inventory and transportation costs. Modeled 70+ trucks across 50+ locations, generating cost-efficient routing solutions.
Technologies Used: IBM ILOG CPLEX (OPL Studio), VBA