Debalina Maiti

Debalina Maiti

An Aspiring Data Enthusiast!

Contact Me

About Me

Debalina is a seasoned product and data professional with solid background in Cloud-based SaaS products and Supply Chain business modules.

She is a data enthusiast and uses data to help companies find insight out of it. What she loves about data is that done right it can help solve some of the world's most challenging problems.

There are always so many new technologies, theories and best practices being developed by the community that new learnings are never in short supply. See below some of her skillsets and she is always looking for more!

When not dealing with data, she is a travel connoisseur and loves to work out.

Applied Machine Learning Projects

NN

Click Through Rate Prediction at Large Scale

In the current landscape of digital advertising, Click Through Rate (CTR) prediction is very important for both advertisers, the entities paying for the ads, and publishers because the cost of the ad is determined by the cost per click. This model will predict CTR in large scale.

Technologies: Spark, Python, Panda, Scikit, SciPy, GCP

Model Used: Logistic Regression

Find out more

NN

News Topic Classification

In this project, text data from newsgroup postings on a variety of topics are extracted. Then model built to distinguish between the topics based on the text of the posts..

Technologies: Python, NumPy, Scikit, Jupyter Notebook

Model Used:K-Nearest Neighbor

Find out more

Check Other ML Projects

Deep Learning Projects

face

Facial Recognition with CNN

The purpose of this project is to predict keypoint positions on face images. This can be used as a building block in several applications, such as: Biometrics / face recognition etc.

Technologies: Python, NumPy, sklearn, lasagne, Theano, AWS

Model Used:Convolutional Neural Network (CNN)

model

Named Entity Recognition (NER) on Clinical Text

Most of the medical practitioners/ clinics use electronic health records to build and record the medical history of patients, which can help enormously in medical. But most of these petabytes of data are in handwritten natural language, making it quite hard to be predicted. In this project annotated clinical data set (provided by i2b2) for recognizing named-entity from raw clinical data. Bidirectional LSTM architecture is being used for NER in clinical text.

Technologies: Python, Panda, SciPy, sklearn, Tensor Flow, AWS

Model Used:CNN, Bidirectional LSTM (Long Short-Term Memory)

Check the abstract

Check my other Deep Learning Works

Android App Building

model

ODYSSEE, an app-based scavenger hunt in Yosemite National Park

Odyssee is focused on a lack of engagement with the natural world affecting over 60% of Americans and urban populations around the globe. Using over 500,000 photos from over 8,000 species, our team fine tuned a nature specific neural net model, applied computer vision and edge technology to create a fascinating game and scavenger hunt experience through National Parks.

Technologies: React Native, Flutter, Javascript, DART, Android Studio, Python, AWS, Google Firebase, Google Analytics, Tensor Flow Mini

Model Used:Vision Model - Inception v3

Odyssee Website

Demo

Internet of Things (IOT)

model

Automatic Sarcasm Detection

Automatic sarcasm detection is a difficult problem for machines because the exact same sentence can be interpreted both literally and sarcastically depending on the context and opinions of the author. While past papers on sarcasm detection informed different modeling approach, in this project recent advances in NLP and utilized HuggingFacePre-Trained BERT [2] Transformer architecture are used to build the classification model.

Technologies: Python, PyTorch, Docker, IBMCloud, JetsonTX2

Model: BERT, Hyper Parameter Tuning using CNN on top of BERT

Architecture
New

Check the Abstract

model

Home Security System using IOT and Edge Device

The objective of this project is to build a lightweight IoT application pipeline with components running both on the edge (your Nvidia Jetson TX2) and the cloud (a VM in Softlayer) to be able to capture faces in a video stream coming from the edge in real time, transmit them to the cloud in real time, and save these faces in the cloud for long term storage.

Technologies: Python, Docker, Mosquitto Broker, IBMCloud, JetsonTX2

Statistical Analysis Projects

model

Crime Trend Analysis

The purpose of this project is to analyze the relationship between crime rate and demographic markers in the city of North Carolina

Technologies: R, Anaconda, Jupyter Notebook

Check the Abstract

model

What factors lead to damaging forest fires?

In some areas Forest Fires are major environmental concern, endangering human lives and causing substantial economic damage. So this amalysis project is motivated by the question: What factors particularly lead to damaging forest fires?

Technologies: R, Jupyter Notebook

Check the Abstract

Data Visualization

model

US Household Debt to Income

The purpose of this project is to visualize the impact of the global financial crisis on income and debt across various states (e.g. rich vs. poor states). The visualization utilizes median household income and debt data for the US and for each US state for the time period of 2003-2018 to understand and compare the financial health of the average household across the US.

Technologies: Tableau, HTML, Javascript, D3

Project Abstract

Find more

More works on DataViz

GitHub Feed

Loading the data just for you.