×
Case Studies Blog Careers Technologies
Computer Vision - ML - Gen AI – AWS Based Intelligent Solar Installation Assessment
Industry
Energy
Technologies
AWS Sagemaker, AWS
About Client

A Leading Renewable Energy Company in India

Problem Statement

As Client rapidly expanded its solar rooftop installation business, ensuring quality and compliance became increasingly challenging. The existing process relied heavily on manual site inspections and photo assessments for warranty approvals, which led to:

  • Long delays in warranty processing.
  • Inconsistent assessments due to subjective human judgment.
  • Limited scalability as business volume grew.
  • Increased operational overhead and bottlenecks.

Client was looking for an automated, scalable, and consistent solution to streamline quality inspections, accelerate approvals, and reduce operational dependencies.

Solution

Oneture partnered with AWS and Client to design and implement an Intelligent Solar Installation Assessment System, blending Computer Vision, OCR, ML and Generative AI technologies.

Key Solution Highlights:

  • Automated Image Analysis: Developed 10+ custom Computer Vision models covering 22 inspection points to assess solar installations against quality standards.
  • Feature Extraction: Leveraged Object Detection, Image Classification, and OCR to extract critical features from uploaded site images.
  • Intelligent Approval Workflow:
    • Auto-approval for images scoring above defined confidence thresholds.
    • Automated notifications for missing or poor-quality images.
    • Escalation to manual review for edge cases.
  • Generative AI: To address complex cases like Distance Measurement where Computer Vision models werent able to predict due to lack of reference in the images, Gen AI models like Segment Anything and others were used to implement the same
  • AWS-Powered Architecture: Fully cloud-native deployment using Amazon SageMaker, Lambda, S3, Ground Truth, and CI/CD pipelines for scalable model development, retraining, and deployment.

Solution Architecture

  • Preprocessing & Completeness Check: Salesforce integrated image capture followed by AWS Lambda-driven preprocessing.
  • Model Training: SageMaker Studio Pipelines with Ground Truth labeled datasets.
  • Continuous Retraining: Automated feedback loop capturing low-confidence predictions for retraining and model improvement.
  • Governance & Reporting: Performance monitoring through precision, recall, F1-score, and processing time KPIs.
Project Execution Approach
  • PoC Phase: 2 months
    • Phase 1: Elevated Component
    • Phase 2: Earthing & Lightning Arrestor
  • Full Project Implementation: 5 months for production-grade system development.
  • Post Deployment Support: Ongoing model monitoring and retraining.
Value Delivered

Key Benefits Delivered

  • 90%+ accuracy in compliance assessment.
  • 60-70% reduction in manual review workload.
  • Significant improvement in warranty approval turnaround time.
  • Highly scalable architecture to support future business growth.
  • Continuous model retraining ensuring system accuracy even as installation practices evolve.
Technologies
  • Amazon SageMaker
  • Amazon S3
  • Amazon Lambda
  • SageMaker Ground Truth
  • CI/CD Pipelines
  • Salesforce Integration
  • Computer Vision (VGG, ResNet, etc.)
  • Generative AI (optional modules)
Risk Management

We proactively addressed potential risks like:

  • Data Bias & Quality Issues: Meticulous dataset curation and continuous feedback loop.
  • Model Drift: Regular retraining and model performance monitoring.
  • Edge Case Handling: Human-in-the-loop workflows for complex scenarios.
  • Security & Explainability: Robust governance and stakeholder training.