This is multiple projects dealing with fraud detection. The client receives data from the point of sales machines (POS) based on manual rules set, the data is transformed for processing. The primary objective is to clean the data and perform features reduction and build machine learning models using H2O AutoML tool and provide the serialised MOJO trained model file for deployment into the POS machines. Read/Research on various ML techniques to improve the data quality and model quality.

• Feature Engineering/Selection using multiple algorithms like CatBoost/RFE/RandomForest etc from scikit learn, identify Correlations, using Weights of Evidence / Information Value (WOE/IV), aggregation of features, outlier detections (Std Dev/ Boxplots).
• Create custom modules, functions, document the processes, evaluate the existing functionalities.
• Missing value Analysis, Feature reduction and imputation

Preprocessing/Cleansing of data extracted from Service Now tool, using NLTK, Scikit Learn, perform feature engineering
Built RNN (LSTM) model to predict appropriate category of Urgency from multiple classes based on the summary/description.
It helped in better classification of urgencies that needed to attend
Achieved an accuracy of 83%. Monitored loss for early stopping, used drop out, and an optimizer to get best results.
Reduced the false alerts by more than 25%
Used Flask microframework for UI.

Successfully planned & designed to migrate data from FalconStor VTL to Data Domain.

End-to-End delivery of the configuration item taken for an upgrade.

Documentation of runbooks and process related documents to enable for a smoother handoff.

Technical discussions on the progress of the project with the onsite team.

Troubleshoot issues by assisting support teams, which eventually can reduce downtime (when time permits).

Implemented EMC Data Domain storage systems at two sites and enabled replication between the two.

Configuration of new backup servers using Veritas NetBackup and inducting them into production.

Configuration of VTLs on the Data Domains and attaching them to the Backup servers.

Ensuring data backups happening between the two and handing off them to the production teams.

• Develop proof of concepts and build solutions as a prototype and evolve the product.
• Design, Model, Implement the development of Machine Learning Operations (MLOps).
• Research and understand the core competencies and tools required to build and deploy a machine learning model into production.
• AWS Sagemaker / DataRobot / Domino Data Labs / MLflow /H2O are some of the tools identified for MLOps and worked on them.
• Support on the documentation of blogs
• Deployment destinations include Web APIs, containerize using Docker

The project is about raw data in the form of text files retrieved from the cameras installed in a parking lot. Each text file contains data about the vacancies in the parking lot, which tells whether a vehicle is there in the slot or not, the timestamps, and the quality of the image inferred from the camera.

• Image data from multiple cameras installed at the parking lot. It contains the image details and their status at different timestamps in a day.
• Analyse the Dashboard data extracted from the IoT devices, which usually contains details about the images captured
• Automate the received data and perform analysis based on the (un)detection rate(good/bad).
• Write python scripts as per the requirement to automate things and export the data into an excel format (as desired by the client).