Milan Kranti

Worked for an American retail & ecommerce company for optimizing a pricing strategy of the products for its presence in US. Performed Data Cleaning and EDAs with data mining algorithm and statistical pattern recognition to understand business-relevant insights including base & coupon prices, sales patterns or trends, promo variations and discounts and make data ready for modeling. Engineered and Created a Python script using Python, Pandas, SQL and Big Query to automate ETL process, reduced manual efforts and saved 10+ hours per week. Designed and Developed an API for post-simulation & actualization using Python and SQL with OOPS concept and automated this using cronjob schedular, reducing the time spent by 90%. Implemented different machine learning algorithm including Regression, Clustering and Random Forest to develop efficient and scalable model and provide simulation results for optimization. Forecasted the 6 months sales at product level by using time series forecasting models and finalized the best model with high accuracy. Helped the team to formulate logic for business related KPI and database schemas for the dashboard. Applied Natural Language Processing analysis to identify similarities between old and new products using faiss vector database.

Worked on a project that helped to understand the different plastic taxes and provided a solution for reducing taxes for products in various countries. Developed and Implemented task and stream to load the data in Snowflake from Azure Data Lake Storage using SQL save daily jobs. Led the design of the pipeline and a Stored procedure using SQL to unload data in Azure Data Lake Storage from Snowflake cloud database and automated this task using Azure Data Factory to increase efficiency by 80%. Implemented Power Bi to visualization tools to understand the data and time series forecasting Dashboarding.

Trained a CNN model for Pizza classification using Python libraries TensorFlow, Keras API, Pandas and NumPy, which can classify pizza as good or burn. Split the whole dataset (2000 images) as 70% training, 10% validation, and 20% test data and use ImageDatagenerator to generate new images in each epoch and achieve 92.57% accuracy. Deploy the above deep learning CNN model on the Android app using the tflite plugin. In a developed Android app, we can take an image from the camera or folder and classify it as a burn or a good pizza.