Implementation of Video Analytics for Retail stores in Middle east. Project involves tracking of Humans and the path followed by them inside the shop. Recognise Objects, humans and track them to obtain data which can be used to derive business insights.

Led the implementation of face based image retrieval for several large Amusement parks and
water parks across Asia. Responsibilities in this role includes: 1. Design the architecture/server
specifications for the implementation of the solution both on premises and on cloud. 2. Decide
the right algorithm to be used for face detection and face embedding calculations keeping in mind
the accuracy and performance requirements. 3. Design a viable servicing strategy to ensure the
application can cater to multiple concurrent requests. First of the kind application to be
implemented on such a large scale is currently hosted on several major tourist attractions in
middle east and can retrieve the images of customers taken at the attraction based on their
scanned face.

Assist in building a data veracity product which helps in figuring out the data quality issues.
Applied ML and Computer Vision to quantify the quality of data and recommends ways to
improve it. Experience of running ML algorithms in a cluster based setup Guide a team of 20 to
scale up in machine learning and deep learning Use CV and CNNs in several OCR and
Classification problems Working with Flask APIs to expose the data pipeline to front end UI.
Details of Projects handled: 1. Fraud Detection using machine learning for a major European
Telco. Skills Applied: Classification Algorithms,Handling imbalanced classes, cross validation
approaches and scores, Dask framework. 2. Anomaly detection on any structured data and
provide recommendations for incomplete and inconsistent records. Skills Applied: Unsupervised
Anomaly detection algorithms , per-processing approaches , SVM, Random forest, Light GBM,
XG Boost regression

Implement fraud detection for a large scale Telecommunication provider.

Features -

Implementation on highly imbalenced dataset

Achieved Hit rate of ~70%

Real time fraud detection solution

Implementation of credit scoring model for enterprise customers.

- Custom solution to arrive at credit score for Enterprise/Corporate Customers.