Sumedh Thanekar Email & Phone Number

Potsdam, Brandenburg, Germany

Frankfurt Am Main, Hesse, Germany

• Worked on a dataset consisting images of around 16,000 different industrial parts• Employed a variety of image pre-processing techniques, such as background removal, artificial background synthesis, and object extraction, for data augmentation• Developed a universal data generator for processing the dataset as per the model requirements• Researched and implemented state-of-the-art techniques in Convolutional Neural Networks and Embedding Networks• Applied EfficientNet and other Embedding Losses like Triplet, Angular and Center Loss, to improve the performance of the object classification model• Used Docker Container to run the models with larger dataset to leverage the GPU cluster• Conducted a thorough evaluation and visualization of the results to gain valuable insights to better communicate the model's accuracy and losses

Rülzheim, Rhineland-Palatinate, Germany

• Worked closely with stakeholders to understand the business needs for deploying an IBM Load Sharing Facility System on ITK GPU Server• Configured the setup of the MLOps pipeline to streamline deployment, monitoring, and management of machine learning models in production• Created and implemented a data validation tool for time series Railway's Dataset, ensuring data quality through careful data preparation and validation processes• Utilized advanced techniques, such as Short Time Fourier transformation, to effectively transform the data and extract relevant features for analysis• Developed a custom data validation feature using the Great Expectations Framework to detect the peak frequency and established data quality metrics• Conducted in-depth analysis of the results to extract meaningful insights, identify trends, and convey the findings to higher management

Rülzheim, Rhineland-Palatinate, Germany

• Conducted research on different Automated Machine Learning (AutoML) Frameworks• Tested these frameworks for time series forecasting and image datasets for class prediction• Configured the hyperparameters of the frameworks to fit the use case and trained the models accordingly • Checked the compatibility of these AutoML libraries with different schedulers like MLFlow and Kubeflow• Visualized and evaluated the results for each framework and reported the findings

Aachen, North Rhine-Westphalia, Germany

• Developed an Image Classifier model using labeled images and trained the model using Azure Custom Vision to detect the faults in a Gasket.• Gathered images by setting and adjusting the camera focal length and background light and capturing the images of different Gaskets by changing their positions. • Deployed the trained Computer Vision model in the Wrist Band Manufacturing Unit at DCC, Aachen, and achieved an accuracy of 85%.• Analyzed the pain points of the factory workers and ideated opportunities by creating a user journey and a website wireframe for the digital capture of problems and their solutions.• Developed a business model to improve the production and attain maximum ROI and reduce the manufacturing loss.

Pune, Maharashtra

• Performed development and testing of a Smart Assistant for a Vehicle Infotainment System using Google Cloud• Worked with the data engineering team to understand the audio data processing requirements for the speech-to-text model• Assisted in designing data pre-processing techniques to reduce noise and enhance voice quality by using a machine learning algorithm to complete the voice command• Built an ETL pipeline that ingested data collected from 5 different IoT sensors of an Autonomous Vehicle and a Regression model to predict the repair time of the vehicle • Assisted in project planning, tracking the progress, and helped identify risks and proposed mitigation strategies• Presented trends and patterns in sensor readings from different environments using Power BI Dashboards, providing valuable insights to stakeholders

Pune, Maharashtra, India

• Developed a predictive maintenance system for the automation of a cement plant using PLC and Energy Analyzer. • Any anomaly in the voltage or current values were analyzed using the PLC and was communicated to the plant in charge via a GSM module. • This data was stored and given as input to a Multivariate Unsupervised Anomaly Detection model built using Isolation Forest.• The model was able to predict plant equipment failure with an accuracy of 85%.Tools and Platform: PyCharm, Excel, ABB Control builder plus, Panel Builder 400

Master'S Degree, Data Analytics And Decision Science, 1.6

Bachelor Of Engineering - Be, Electrical, Electronics And Communications Engineering