Asset Inspection Saves Lives

Deployed backend service to prevent dangerous site visits

Project context: External blogpost

Summary

Personal contributions:

  • Reduced health risks by automating site inspection with a satellite computer vision product deployed in GCP Cloud Run;
  • Achieved acceptance criteria for enterprise ML project, allowing on-time project delivery, by successfully negotiating error-correction web-app feature with client;
  • Developed Explainable AI (XAI) SHAP solution for revenue drivers for a gas-industry client;

Situation

The project’s success depended on the rapid and reliable detection of anomalies on floating tank sites using satellite imagery. Two primary challenges emerged:

  • Safety and Efficiency: Traditional, manual site inspections posed inherent health and safety risks to personnel and were time-consuming.
  • Client Acceptance: Meeting the enterprise client’s strict acceptance criteria required a high level of accuracy and a clear process for handling model errors, which threatened the project’s delivery timeline.

Task

My core responsibilities were to operationalize the model for maximum safety and impact, secure final project acceptance, and equip the client with tools to understand the model’s decisions.

The specific tasks included:

  • Developing and deploying the model into a resilient, scalable, production environment to automate site inspection.
  • Negotiating and developing a solution that addressed the client’s need for error correction to meet acceptance criteria.
  • Implementing Explainable AI (XAI) to justify the model’s outputs and build client trust.

Action

I executed three critical actions that ensured project success and expanded its value proposition:

A. Automating Site Inspection for Safety

I led the effort to move the satellite Computer Vision product from development into production. This involved:

  • Developing the necessary deployment structure to host the model.
  • Deploying the solution on GCP Cloud Run, utilizing its serverless container capabilities for scalable and cost-effective inference. This automation successfully reduced health risks by eliminating the need for personnel to manually inspect hazardous sites, replacing it with a reliable, remote solution.

B. Ensuring Client Acceptance with an Error-Correction Feature

Recognizing that model output, while highly accurate, would still contain false positives or negatives, I proactively addressed client concerns regarding data trust and corrective action.

  • I successfully negotiated an error-correction web-app feature with the client, which allowed their operations team to review, validate, and correct the model’s predictions directly.
  • By delivering this feature, we provided the client with necessary control and oversight, which was instrumental in achieving the acceptance criteria for the enterprise ML project, allowing for on-time project delivery.

C. Developing SHAP for Revenue Drivers (Cross-Project Value)

Leveraging my expertise across projects to demonstrate value, I concurrently developed an Explainable AI solution to provide broader strategic insight.

  • I utilized the SHAP (SHapley Additive exPlanations) framework to explain the drivers behind revenue models for a separate gas-industry client.
  • While applied to a different domain, this action demonstrated the team’s capability to provide Explainable AI (XAI) beyond simple model predictions, fostering a culture of transparency and data-driven strategy.

Result

My contributions led to a more robust, safer, and client-acceptable solution:

  • Reduced Health Risks: The automated site inspection product, deployed in GCP Cloud Run, successfully reduced health risks by automating the hazardous site inspection process.
  • On-Time Project Delivery: By successfully negotiating and implementing the error-correction web-app feature with the client, we achieved acceptance criteria for the enterprise ML project, allowing on-time project delivery.
  • Enhanced Strategic Insight: The development of the Explainable AI (XAI) SHAP solution for revenue drivers demonstrated an advanced capability to deliver model transparency and strategic insight to gas-industry clients.