The predictive quality of the rod weights of cigarettes to ensure that the manufactured cigarettes meet the specific quality requirements set by the manufacturer.
• Designed the approach and architecture of the solution.
• Define the data gathering and analysis process with domain experts.
• Analyzed a large amount of sensor data, determined features important for modelling historized the sensors with the help of the Data Engineering team.
• Provided weekly analysis report helping the client to understand the information obtained from the data and take actions concerning their current operations.
• Created Power BI dashboards to articulate the findings in a visually appealing way.
• Created data preprocessing pipelines to process enormous sensor data in Databricks. Cleaned and merged data coming from different sources making it suitable to apply ML.
• Defined logic to merge the data with the help of Data Engineers and Solutions Architect.
• Currently developing ML and DL based algorithms to intelligently monitor the quality of the product and generate alerts if there is any degradation in the quality.

• Created microservices as Flask APIs and deployed them using docker container on Azure web apps.
• Built predictive models on manufacturing data to help the company make data-driven decisions.
• Built anomaly detection models to identify anomalies in sensor data coming from manufacturing bed and send alerts if an anomaly is detected in any sensor values.
• Built intelligence around anomaly models to suggest correct desired levels of sensor values if an anomaly is detected to maintain continuous operation of machines.
• Helped in creating Azure IoT Hub to collect and process raw sensor values and store them in the desired format in Azure blob storage.
• Used Azure Machine Learning services to quickly build and test various machine learning models such as SVM, Random forest, XGBoost, LightGBM etc.

Semi Automatic Image Annotation Toolbox with RetinaNet, SSD and YOLO as the suggesting algorithm. We can select from the three algorithms as the suggesting algorithm. Filters can be applied to images to create distorting images which will help in the training of the model. Also the training process is incoperated in the application. You can train a Retinanet model for custom objects in one click.