Athira Das

An online self paced course rooted in adult developmet theory for change management in SMBs for AI adoption

This paper examines the transformation in workplace dynamics precipitated by Artificial Intelligence (AI) advancements, particularly focusing on the shift from a Thinking Economy to a Feeling Economy.

This paper explores the feasibility and design of an AI-powered coach using Large Language Models (LLMs) in the context of organizational coaching. The first part is an exploration of the current state of LLM-based models. The second part evaluates an LLM-based AI model for organizational coaching using Terblanche's Designing AI Coach (DAIC) framework. A conceptual chatbot design using LLM-based models that can be used for organizational coaching is proposed in the paper.

Tools used: Tableu

Analyse the interaction between individuals on SAP knowledge sharing platform. The analysis were based on centrality measures and community formation.

Tools Used: R, Gephi

The customers were segmented based on purchase behaviour for Target Marketing.

Different algorithms like k-means, k-medoid, agglomerative clustering and dbscan clustering, K-means which had the best separation based was selected to segment the data. The model was validated with decision tree.

Develop a credit scoring rule that can be used to help determine whether a new applicant presents a good or bad credit risk.This was done as a part of IDS 572 - Data Mining course at University of Ilinois at Chicago.

Data Used: German Credit data
Model: Decision Tree

Analytics based rule engine that effectively sifts through Billing information of customers to spew out potential errors, catches error patterns & calculates revenue leakages. Enables Retail energy providers with 100% accurate bills before being sent out to customers.

 Developed an automated system for downloading the data from external sources using Python.
 Developed the ETL process for transforming the data using python.
 Developed the rules engine for billing assurance using… Show more

Analytics based rule engine that effectively sifts through Billing information of customers to spew out potential errors, catches error patterns & calculates revenue leakages. Enables Retail energy providers with 100% accurate bills before being sent out to customers.

 Developed an automated system for downloading the data from external sources using Python.
 Developed the ETL process for transforming the data using python.
 Developed the rules engine for billing assurance using SQL procedures.
 Designed and developed the interactive user interface. Show less

Customer Micro-segmentation an application which classifies the consumers with respect to their financial value to the company, timeliness of payment and their loyalty towards contract renewals

 Generated the EDAs for customer behaviour analysis.
 Ideated and developed the important KPIs.
 Developed the ETL process for data transformation.

Cerebra signal studio is Flutura’s flagship platform to collect the real time and batch data, perform analytics and show the result through creative visualizations.

 Ideated and developed the important KPIs.
 Developed the predictor ranking algorithm for predicting the important parameters.
 Developed the template engine for the vectorization process using Python.
 Developed the ETL process using Apache Spark and SparkSQL.

The project is based on applying supply chain forecasting concepts to the deregulated electrical energy market in New York State. We will attempt to forecast the energy demand and pricing for New York State and each of their 11 zones for three years. We will train the data using the last 4 years of data and determine how accurately our forecast matched the actual demand. Since the NY Independent System Operator also posts their forecast demand, we will additionally compare to their initial… Show more

The project is based on applying supply chain forecasting concepts to the deregulated electrical energy market in New York State. We will attempt to forecast the energy demand and pricing for New York State and each of their 11 zones for three years. We will train the data using the last 4 years of data and determine how accurately our forecast matched the actual demand. Since the NY Independent System Operator also posts their forecast demand, we will additionally compare to their initial forecast. The forecasted price will be used for the procurement planning for the Retail Energy Provider.
Show less

This problem is around image recognition and multi-label classification. Here the tag doesn’t indicate the object in the image, but rather a concept represented by the restaurant. The concept of transfer learning helped us to build the model with more efficiency. After image augmentation, we used Google’s Inception V3 to convert the images to a numpy array. Later this feature matrix was used to build a model using XGBoost, then created a model which was a combination of XGBoost and clustering… Show more

This problem is around image recognition and multi-label classification. Here the tag doesn’t indicate the object in the image, but rather a concept represented by the restaurant. The concept of transfer learning helped us to build the model with more efficiency. After image augmentation, we used Google’s Inception V3 to convert the images to a numpy array. Later this feature matrix was used to build a model using XGBoost, then created a model which was a combination of XGBoost and clustering following by an ensemble of the two using a custom Neural Network Model. We tested the efficiency of our model using F-score. Show less

Develop a data mining model to improve the cost- effectiveness of the direct marketing campaign of Paralyzed Veterans of America.

Data Used: The dataset used KDD cup 1998, http://kdd.ics.uci.edu/databases/kddcup98/epsilon_mirror/cup98dic.txt
The data was analyzed to predict the users who are going to make donations for the coming year. The model was developed to predict the rare event - "User who can donate". The final model was selected based on combining the response and donation… Show more

Develop a data mining model to improve the cost- effectiveness of the direct marketing campaign of Paralyzed Veterans of America.

Data Used: The dataset used KDD cup 1998, http://kdd.ics.uci.edu/databases/kddcup98/epsilon_mirror/cup98dic.txt
The data was analyzed to predict the users who are going to make donations for the coming year. The model was developed to predict the rare event - "User who can donate". The final model was selected based on combining the response and donation amount models to
identify the most profitable individuals to target.

Models used for Classification: Decision Tree, Naive Bayes, Logistic Regression, Random Forest, Boosted Trees, Support Vector Machines
Model used for Regression: Linear regression.


Show less

Cerebra Data Products are a set of 30+ products made on a common platform with a unified data model for power utility industry catering to generation, transmission and distribution segments.

 Designed the data model for the utility industry.
 Ideated and developed important KPIs for the utility industry.
 Built a real time data pipeline to store data from multiple sources, for real time analysis and Batch Processing using Apache Kafka, Apache storm, HDFS, Hive, Pig and Cassandra.

The competition was hosted in Kaggle.
Finding a mathematical model to increase the effectiveness of investments in new restaurant sites would allow TFI to invest more in other important business areas, like sustainability, innovation, and training for new employees. Using demographic, real estate, and commercial data, this competition challenges you to predict the annual restaurant sales of 100,000 regional locations.