Sakshi Atrik

Duplicate Warranty Claim:- • Designed a data processing pipeline to identify and flag duplicate warranty images using metadata-driven filtering. • Built automated workflows using AWS Lambda and S3 to ingest and preprocess high volumes of image data. • Stored image metadata and matching results in DynamoDB to enable fast lookups and reduce redundant data processing. • Integrated the pipeline with Salesforce (SFDC) to report duplicate claims and support automated decision-making. • Improved warranty claim data quality by detecting and filtering duplicate images before downstream processing. 2. Marketing Operating Price Analysis:- • Designed and maintained an end-to-end data processing pipeline to extract, validate, and store invoice data from warranty claim submissions. • Built serverless ingestion workflows using AWS Lambda, S3, and EventBridge to process invoice files in near real-time. • Implemented document preprocessing pipelines (PDF → Image conversion) to standardize unstructured inputs and improve OCR reliability. • Integrated Azure OpenAI–based invoice value extraction into the pipeline with robust error handling and retry logic. • Posted validated results to Salesforce (SFDC) for downstream workflows. • Processed ~8,000 invoice documents per day using an event-driven architecture. 3. Sub Dealer Board:- • Developed a scalable data pipeline to ingest and process survey data from sub-dealers across multiple regions. • Implemented batch processing workflows to handle periodic survey uploads with reliable error handling. mailto:sakshiatrik@gmail.com http://www.linkedin.com/in/sakshi-atrik • Stored cleaned and structured survey datasets in cloud storage to support reporting and operational insights. 4. Defect Claim Image Processing :- • Built an event-driven image processing pipeline to identify tyre defects from warranty claim images. • Deployed a YOLOv11-based defect detection model on Amazon SageMaker and exposed it via a real-time inference endpoint. • Triggered inference using AWS Lambda on S3 image uploads. • Processed model predictions and stored structured defect metadata in DynamoDB using the image filename as the partition key. • Enabled downstream systems to consume defect classification results through metadata-driven storage and APIs.