Quant Data Science

Auction Parameter Optimization: Highly Rated Search Thematic Widget• Examined the operational workflow of sourcing, ranking and display of Sponsored Products ads on user search pages• Tuned the parameters of auction scoring function for improved ad ranking in Highly Rated search thematic widget• Employed sobol search strategy for parameter optimization, enhancing long-term shopper metrics by 40% in IN marketplaceAdvertisement-Query Relevance model using LLMs• Developing an ad-query relevance model via implementation of Parameter-Efficient Fine-Tuning techniques on LLMs

• Built a robust feature extraction pipeline with three core components: expansion, selection and dimensionality reduction• Explored python libraries for automated feature expansion of time series data in order to maximise information content• Delved into feature selection algorithms like RFE, BorutaSHAP to drop out irrelevant features from the expanded set• Leveraged PCA and Autoencoders to capture maximum variability within the chosen features into a compact feature set• Evaluated using OLHCV data for monthly stock price trend prediction with LSTM achieving an accuracy of 0.85

• Examined Planetary Computer for extraction of Sentinel-2 images for a given Area of Interest (AOI) and time of interest• Selected the satellite image with the highest coverage area from the set of output images returned for a given input query• Used Scene Classification Layer (SCL) band for masking clouds, masking the pixels with band values as 0, 8, 9, 10• Applied morphological operations like dilation & erosion using OpenCV over the stored .tif files of RGB and NDVI bands

• Explored document understanding models like LayoutLMv2 and LayoutLMv3 for extraction of entities from mortgage documents• Improved F1 score of 0.76 for a document dataset, surpassing the existing NER approach which had a score of 0.66• Analysed the results of LMv3 on the document dataset with that of the existing results obtained using LMv2• Explored models like LayoutXLM to build key-value relation extractor for various form documents and wrote inference scripts for Semantic Entity Recognition of form documents for actual deployment to the production server

Awarded Research Grant for working under the supervision of Prof. Salil Goel on the topic: "Object Detection on LiDAR Image" under the SURGE program 2022. • Designed an object detection algorithm in Tensorflow to detect two classes on a dataset containing images collected from highways and rural areas• Analysed the KITTI dataset to perform detection and 3D localization of two classes: Car & Pedestrian• Used PyTorch to develop a Convolution Neural Network that takes LiDAR point cloud and images as input simultaneously

• Acted as a point of contact for over 10 companies during the Placement Season as well as in the Internship Season 2021-22• Worked with a 3 tier SPO team and managed interviews, Pre-Placement Talks, Group Discussions and tests for about 800+ students during the Internship Season and about 1200+ students in Placement Season 2021-22

• Collaborated with a team of 45 AI engineers to build an automated Plant Health Prediction Model.• Used QGIS to analyse the dataset comprising of drone-derived multispectral and thermal data combined with irrigation data.• Identified the concerned regions with water stress, pest infestation, man-made plant stress and created shapefiles for the same.• Impact: Improved Irrigation Management to save and optimize water consumption as well as to reduce the use of pesticides.• Invited to join the Omdena Community. Invitations are sent to the top performers in the challenge.

• Mentored various students in UnchaAI , an ed-tech startup by IITK and NITP alumni which provides a mentorship platform to JEE aspirants.• Guided JEE aspirants with their PCM doubts and designed a customized time table & metric to track their performance periodically.• Conducted regular motivation sessions which helped them in gaining self confidence during the pandemic period.