sarthak agarwal

Propensity Module: Module consisted of multiple parts Cohort Creation, saving cohort meta data, saving users details laying in respective cohort, generation of suitability matrix and recommendation generation using propensity scoring. Implemented Tree and Graph based approach for storing the historical dataset for reducing time complexity in cohort generation, handling of large dataset with row count of approx 1cr. Error Logger: Implemented error tracking on servers and sending the traceback to respective project owner on slack using 2 approaches python watchdog and handling stdout and stderr streams providing one approach without developer intervention and one for controlled and specific error tracking. Reconciliation Module: Worked on reconciliation project for commission consisting of grid upload, rules generation for reconciliation of commission calculations on multiple types of grid and automatic generation of mis report. Make Model Mapping: Worked on make model variant mapping between insurer data and Zopper's internal data for 4 wheeler, 2 wheeler and miscellaneous vehicles using jacard similarity and n-gram combined approach and adding some rules and data modification using token set and token sort ratio.

Drawing insights from a large database on realization i.e. conversion of online booking on oyo platform to people actually checking in hotels. Development and maintenance of multiple trackers for realization team.

Worked on various projects like Face Extraction and Face Verification using AWS Recognition, Watermark Image Classification and Watermark Removal using Resnet50 and G.A.N neural networks, KYC Validation. Basic Tech stack included Python, Flask, FastAPI, PostgreSQL, neural networks, machine learning algorithms, Docker containers and Docker Compose. Worked on refactoring and reforming previous build features of under development products. Handled the responsibility of end to end production and deployment of various features. Worked on data collection, data labeling, data annotations and model training, provided support and maintenance for various per-existing services. Worked on project A.N.P.R. (Automatic Number Plate Recognition) where we tried to detect the number plate in an image of a car and extract license number and number plate color. Tech stack used was Python, FastApi, Docker, Unet neural network, LabellImg labeling tool, AWS Textract.