VERNIKA GUPTA

Total Experience: 5+ Years

Data Science Experience: 4+ Years

Accomplished Data Scientist skilled in translating complex business challenges into actionable insights. Expertise in advanced Image Processing for precise OCR and refining cutting-edge deep learning models. Strong aptitude for machine learning, predictive analytics, and efficient cloud resource utilization. Proficient in Python, Databricks, and AWS, poised to drive impactful solutions for enhanced business outcomes.

Project: Time Series Modelling for Profit and Loss Statement Prediction in Different Segments: Driver-Based Statistical Approach

• Employed driver-based statistical approach to identify and rank key drivers impacting financial performance.
• Successfully predicted P&L statements in different segments using univariate and multivariate models.
• Achieved average 90% accuracy among all PnL.

Project: Immunization Volume Prediction:

• Developed time series models to predict immunization volume accurately.
• Improved resource planning and allocation by providing reliable forecasts.
• Deployed the system in Anaplan, providing real-time access and actionable insights to stakeholders.
• Achieved 92% accuracy on an average for different Vaccines.

Project: Automation of Gas Plumes Exploration

The project resulted in automating the processing of 900,000 images and identifying a targeted region with a 95% chance of finding gas plumes within a 100-meter radius. This automation significantly improved efficiency in gas exploration activities.

• Utilized deep learning techniques to classify and detect gas plumes with 95% accuracy.
• Processed approximately 900,000 images of multibeam backscatter and bathymetry data.
• Developed code using fast.ai for plume and non-plume classification.
• Implemented object detection using YOLO to locate gas plumes.
• Created a pixel-to-global coordinate conversion and distance calculation system.
• Designed a business metric for error analysis and extended the solution to multiple locations.

Project: Defect Identification for a CPG Production Plant

Implemented a cutting-edge defect detection system for a consumer-packaged goods (CPG) production plant, leveraging live images from the production line of feminine hygiene products. Developed highly accurate Artificial Intelligence-based Computer Vision models with approximately 98% accuracy, significantly reducing manual effort in defect identification.

• Developed a multi-label classification model using Keras to accurately classify defects in real-time production line images of feminine hygiene products.
• Implemented object detection models, including RetinaNet and YOLO, to identify and localize defects with precision.
• Improved model accuracy by implementing advanced techniques such as data augmentation and hyperparameter tuning.

Project: Signature Detection Component

I spearheaded the conception and execution of a signature detection component meticulously crafted to discern signatures within both unstructured and structured documents. Essential facets of this endeavor encompass:

• Implemented signature detection using OpenCV for structured documents, ensuring accurate identification.
• Utilized TensorFlow object detection to detect signatures in unstructured documents, leveraging deep learning techniques.
• Iteratively improved the model to reduce false positives and enhance detection precision.
• Built an end-to-end solution that seamlessly integrates with existing products or systems.
• Ensured the component is scalable, efficient, and compatible with various document types.

Project: Handwriting Text Extraction
The Cognitive machine reader product suite processes documents and returns extracted information back to customers. The inclusion of text extraction models in the platform enables customers to have a unified solution that can process all variations of data points in a document such as printed, Handwritten, Checkbox, Signature, bar code, stamp. Apache MXNET was the library running underneath for handwriting text extraction.

• Implemented deep learning techniques to extract handwritten data from structured documents.
• Signature and stamp detection was developed on Object Detection with Resnet architecture.
• Built a robust solution that successfully extracted data from documents, achieving a document-level accuracy of 65%.