• Developed machine learning models using statistical methods, RNNs, and BERT to identify core intent-bearing tokens in eCommerce search queries, achieving a 35.41% improvement in F1 score over baseline.• Executed use-case specific analysis for recovery module, improving search results for null and low search query scenarios, achieving a 23% increase in recall and 11.9% increase in relevance over the current production model.• Submitted paper to CIKM Conference 2024 on how query reformulations help identify intent-bearing tokens.

• Redesigning explainable AI (XAI) techniques like saliency maps, CAM methods, D-Rise, forward selection and evolutionary algorithm etc. to rank feature maps of object detection networks to reflect their contribution towards prediction of uncertainty.• Concluding most vulnerable and sensitive classes in Automated Driving dataset using uncertainty and XAI output.

• Created ETL pipelines using SQL and Pandas for NuImages dataset to train object detection models.Researched and implemented object based uncertainty prediction methods by ensembling deterministic deep learning models (trained using Tensorflow object detection API and YOLO), BSAS clustering and Test Time Augmentation.• Performed Knowledge distillation from ensemble to a single model using Ensemble Distillation (ED) and Ensemble Distribution Distillation (EDD) to predict pixel level uncertainty efficiently.

Enhancing web application for the evaluation of a company’s CDR dimensions (eg. Access, Governance, Security etc.) by analysing their website. The Project is funded by Hessian Ministry for Digital Strategy and Development. Worked on improving the frontend application with vue.js. Researched on redesigning analyser code at the backend server to deploy machine learning based NLP methods.

Design of data-driven approaches for predicting milling-tool wear and estimating remaining use-ful lifetime (RUL) using time series analysisDevelopment and implementation of neural networks and traditional machine learning methods for inline tool condition monitoring of an industrial machine using various sensor signals and experimentation dataPrepare implementation of the developed solution into an IT infrastructureSub-tasks include exploratory data analysis, data preparation, feature engineering, feature selection, model selection, and literature research

Created Telemetry Data Decoding and Visualization utility in both MATLAB and Python to assess the trial flight performance which improved the assessment speed by 3 times and Baud rate mismatch, Sensors activity malfuntion etc. are detected and notified to control center in real time.Debugged existing code for tilt detection and devised new algorithm in Embedded C using ’Capture-Compare-Pulse’ module which reduced False positive rate (FPR) by 50 %

Built Standalone application for Surveillance using Image enhancement, Filtering, Colour detection and Object detection etc. using MATLAB, MATLAB GUI and Raspberry pi. Achieved ambiguous activity detection accuracy of 84 %.

Over-viewed the communication systems with practical onsite exposure to field activities including splicing of optical fibers, troubleshooting telephone exchange errors and setting up new data links using Layer 2, Layer 3 switches as well as learnt use of communication system in power utility in generation, transmission and distribution. I learnt use of SCADA in controlling transmission of TATA power network and PLC in distributed control system(DCS) at TATA Power thermal generation plant, Trombay in Mumbai.