Pranjal Soni Pranjal-Soni

Data Scientist (August 2022 - Currently)
- Build an automated Demand Engine for demand forecasting engine for FMCG product.
- Developed a range-based forecasting model with the help of residual bootstrapping to reduce the negative bias and improve the accuracy of the forecast.
- Performed data cleaning and exploratory data analysis and worked on classical series models, i.e. ARIMA, SARIMA, Prophet, and machine learning algorithms, i.e. Random Forest, XGB, LightGBM.
Data Scientist (August 2021 - August 2022)
- Worked on digital signals, filtering, analyzing, and preparing the data. Applied various statistical methods to generate insights and model the signals
- Explicitly worked with python libraries like Pandas, Matplotlib, NumPy etc., to prepare, generate reports, and build dashboards using PowerBI to deliver business insights of the data
- Setup PostgreSQL database on the server, create pipelines for automation to dump the data, and performed data collection
Data Science Intern (6 months)
- Built deep learning models for computer vision project using the Pytorch framework for recognizing text from the images
- Worked with transfer learning technique and used pre-trained models like Resnet, Efficientnet, Nfnets and attention models
- Extract business keywords from articles using collocations and the RAKE algorithm
- Performed data collection, preprocessing and labeling. Improve running time of the python scripts to enhance efficiency of the programs

Repo	Description	Link
Fake Job Description Prediction	End-to-End project to identify the fradulent job descrioptions.	Link
Bitcoin Price Dependency on Stock Market	Test various hypothesis to analyse How Bitcoin price movements are related to or affected by the various fluctuations in the Stock market	Link
Bank Customer Churn Prediction	End-to-End project to predict the probability of a bank customer is going to exit from the bank or not	Link

Repo	Description	Link
Fine Grained Sentiment Analysis of Mobile Reviews	Scrap the mobile reviews from amazon website and trained machine learning model using stanford tree sentiment data. Then predict the sentiments for the mobile reviews.	Link
Frequent Entities Extraction and Sentiment Analysis of Tweets	Identifying the higly frequent entities and predicting sentiment of tweets for those entities using pretraied BERT model	Link
Topic Modelling Of Medium Articles Using LDA	Perform topic modelling of mediem blogs to divide the different categories according to the topics. LDA algorithm is used to perform the task.	Link

Computer Vision Repo

Repo	Description	Link
Dog Breed Image Classification	A end to end computer vision project to idendify the breed of dogs. Resnet50 model is trained on the dog images and the model is able to identify 10 different dog breeds.	Link