Abhijit Purru
Developer at Mobirey- Claim this Profile
Click to upgrade to our gold package
for the full feature experience.
Topline Score
Bio
Credentials
-
AWS Certified Developer – Associate
Amazon Web Services (AWS)Jun, 2021- Nov, 2024
Experience
-
MOBIREY
-
United States
-
IT Services and IT Consulting
-
1 - 100 Employee
-
Developer
-
Jun 2022 - Present
• This project deals with creating a dashboard for multiple crypto currency exchanges • Implemented WebSocket calls with staggered data collection to decrease the amount of data to process • Created and connected MongoDB database to store the collected asks and bids updates for each currency • Cleaned the dataset using Pandas in order to remove unnecessary data points that will skew the data • Plotted the data in real time using Flask on a ticker graph in order to visualize the data trends • Tested ML algorithms using Scikit Learn to predict the volatility of the future projections of the currency • The achieved predictions are in the error range of only 15% from the true value • Rigged Raspberry Pi w/ camera and created dataset of human faces and masks • Labelled acquired dataset using LabelImg with three categories: mask, no mask and mask worn incorrectly • Used custom YOLOv5 model to train dataset in order to predict the correct category of data • Achieved 92% accuracy at detecting the right category of images and videos using YOLOV5 model • Deployed project using Flask and Docker to make useable interface to train and predict images/videos • This project entailed automatic invoice information detection for invoices stored in picture format • Conducted pre-processing steps using OpenCV such as skew-correction, morphology and denoise • Implemented Faster R-CNN and Mask R-CNN to classify data types in any given invoice such as date and tables • Trained images using custom YOLOV5 model to detect types of tables (bordered, semi-bordered and borderless) • Created a new deep learning model that achieves edge detection on borderless tables as they as the most difficult to digitize • Used NLTK vocabulary libraries to find and fix possible word and number detection errors present in the output • Used a deep learning model called layoutlm to further optimize and correct the previous model • Trained models have an accuracy of 86% for detecting the correct information Show less
-
-
-
-
Data Scientist
-
Jan 2022 - Jun 2022
· The goal of the project is to predict, from the chosen parameters, whether the respondent will have a heart attack · Cleaned dataset using Pandas and visualized graphs in order to find the ideal parameters to be used in the models · Added Min-Max Scaler to standardize unbalanced data from the dataset · Used hyperparameter tuning in conjunction with Optimal thresholding to increase the true positives rates · The models that were used are: KNN, Logistic Regression, Random Forest, Decision Tree · Achieved 92% accuracy using Logistic Regression Classification · Created a website using Flask that tracks the amount of COVID cases in the US (updated daily) · Stored the data using PostgreSQL and connected the data to Flask using SQLAlchemy · Implemented bar charts and histograms using HTML, CSS and Leaflet to display COVID total cases, recoveries, and deaths per filter · Created an interactive bubble map using Leaflet to display COVID information by state · Utilized Scikit-Learn to generate predictive models for the spread of COVID in the US over the course of 2021 Show less
-
-
Education
-
2021 - 2023
Drexel University
Master's degree, Data Science -
2015 - 2019
NYU Tandon School of Engineering
Bachelor's degree, Mechanical Engineering
Community
