- Object detection models like Faster-RCNN, RetinaNet, Mask-RCNN, Cascade-RCNN in Tensorflow to detect various telecom equipment with an average accuracy of 88%, enhancing the solution relying more on augmentations, hyper-parameter tuning, anchor optimization, implementing semi-supervised and unsupervised object detection and classification techniques like contrastive learning, DETReg etc. to get good results with lesser labeled data and reducing labeling cost.- End-to-end Machine Learning models deployment and testing to development and production servers – building model, creating Flask service as wrapper for interaction with other micro-services, creating docker containers and scaling these containers with Nginx on GPU machines, implementing CI/CD pipeline. Four ML models has been successfully deployed on production.- CORD-19 Research Challenge: implemented a Search Engine based on BM25Plus, LDA Topic Modelling to let researchers sift through topics and see papers from topic they are most interested in, text pre-processing, lexical parsing, Semantic Tagging, SQuAD BERT, Gensim word2vec model, BERT summarizer, knowledge graphs, Ranked Articles based on external journal list.- Video Intelligence/Relevant Information Extraction project to analyze vidoes, converting them to text and training DistilBertTokenizer and DistilBertForQuestionAnswering model from DistilBERT on the text to get answers to relevant questions.- P3 analysis: Network usage data to gather performance insights for various service providers, time-series analysis, time-series forecast and KPI prediction using regression models, LSTM model, clustering algorithms to figure out areas of varied performance..- RFNet data: Transfer Learning project using LSTM model to predict KPIs for RFNet data trained on one area and identify KPIs performance.

Working as Data Science Intern with the Analytics and Automation Development team developing and providing solution s for the tool EOS 3.0 in Node.js, Polymer, MongoDB, Kubernetes & Dockers and delivering incremental value• Prediction system in R to quote the cost to the customer whenever a new project comes. Simplified a time-taking process calculating the cost based on different project parameters using multivariate linear regression model and rigorous data cleaning and pre-processing. Implemented multinomial Neural Network to predict using softmax function to predict probabilities and align them to output values from user inputs, benefitted 75% cost in this process to the team.• Created time-keeping application tool for the engineers for efficient time-keeping in C#, ADO.Net, using FileMaker as the Database, reducing their efforts for timekeeping by 90%. Testing the tool first using units test cases, then integration, system and performance testing for C#.Secured third place in Innovation Hackathon in North America region. Built the “Luv your Timesheet” application in Microsoft PowerApps and Microsoft Flow and using Tableau for creating report views.• Developed web interfacing tool for reporting in web2py python framework for engineers to pull project site reports for customer, maintaining MySQL database, letting users feed in images, data, select kind of reports using dropdown boxes in UI. Deal with python pptx along with different placeholders to create downloadable pptx site reports.• Created a tool in java being back-end logic language and using Spring framework for front-end to provide desktop interface. It measures cable and other losses while device testing.• Worked on internal web portal implementation using HTML, C#, PHP, JavaScript, J Query, AJAX etc.• Publish Dashboard with important analysis using Tableau.

UNIFIED INVENTORY INTERFACE - UII is central Network and Service Inventory system which based on ODI, OWB, WebLogic, UNIX, Java and PLSQL/SQL. UII is a highly business critical and huge inventory application handles entire wholesale data for British Telecom providing association of network and services data which helps in several functions in Network Planning and Operations like End to End Provisioning, Surveillance, Fault Management, Test & diagnostics, Planning and Capacity Management. • Designed the new interface environment for various interacting components using Apache Hive, PL/SQL, Java, Unix. It improved efficient data updation by 80% and 70% less faults detection.• Implemented various user stories by mapping configurations for all the major ETL object references for the application using Oracle Warehouse Builder 10g. This setup enable us to handle 80% of our customer data.• Setup the secondary site (servers setup) as disaster recovery for UII application consisting of 13 servers – Java war builds, database and other connectivity like sftp, http, jms,dblink etc., environment of the server using unix shell scripting. Reduced client’s revenue loss.• Implemented user interface in Java, Spring framework for this application to be able to handle millions of data requests at a time and enables user’s access to 1000s of relevant records information in a go. It was helpful to site engineers and I was appreciated by BT as well as Tech Mahindra for successful implementation of this interface.• Connectivity setups – ftp, dblink, JMS and http. Implementing useful functionality via scripts through password-less sftp, expertise in authentication key concepts for servers.