Image recognition project where we predicted the lithogolical composition of the sample. This is inteded to be used in Natural resource availability prediction in Oil and gas industry

Designed an Artificial intelligence based chatbot to mimic doctor patient conversation. Used Keras and Theanos to create this chat bot. Bot was trained on 8000 conversatation data and it wias being trained for a week in order to achieve 70% accuracy

Designed an algorithm to extract social influencers of your brand page and create custom posts basis on the topics liked by followers of these influencers.

This analytic helped our customers to reducer their campaign expences by 20% and more.

This was developed for Hilton Hotels. There are 2000+ facilities and all of them could run separate campaign simltaneously.

We designed and developed and algorithm which generated top 5 best suited campaign for each customer based on their Geography, Theme of the campaign and booking propensity of the customer.

These campaigns were helping us to generate leads worth 6 Billion USD in a year

Background: Detect Providers that create nexus and operate in a closed network specifically targeting members living in trailer park

Methodology: Graph pattern mining

* Claims transactional data is converted into an edge list/co occurance graph data

* Each vertex is a provider, and edge between two providers is the frequency of the provider pair fot the same members

*Clustering is done on graph data using "Clique" algorithm. A clique of graph is induced subgraph that identifies vertices/provider that are closely related connected to pne another. i,e. narrow exclusive groups nexus

Tools Used: Python

Output:

* Algorithm identified 280 providers with clique size >8 with a saving opportunity of 138 Million USD

Business Problem: Promo are 10/15 seconds ad where Television channel promote their own contents example " Watch Blockbuster movie on weekend". Our client wanted to minimize these promotions and simultaneousle increase the viewersip for each of promoted programs

Methodology: Designed a Mathematical model to estimatetv viewershipat any time slot, which is calcualted on heuristic model based on simulated output of statistical models.

Tools used: Julia and Python Programing

Output : This was 30% more effecient than their existing manual process

Created analytical solution whcih could identify the adaptive pattern of fraudulent providers, This algorithm was based on Genetic algorithm and classification model.

deployed these to production with annual saving opportunity worth 21Million USD