Projects

Here are some of the recent projects that I've worked on and am currently working on

Development of an AI-driven SOP Q&A Dashboard

Developed an AI-driven SOP Q&A Dashboard leveraging ChatGPT-3.5 and text-embedding-ada-002 for real-time, accurate SOP inquiry responses, enhancing design innovation and human-AI co-creativity.
Integrated system with Azure OpenAI Service for scalability, ensuring robust handling of diverse SOP-related queries across engineering domains.
Implemented advanced analytics to track user interactions and system performance, facilitating continuous improvement and user experience enhancement.
Worked on different use cases such as Code comparison, code generation based on prompts using the same api.

Forecasting sales for an automobile company

• The business case was to deliver monthly predictions of overall 2-Wheeler sales for all area offices in India considering several event variables. Gathered retail and dispatch data for the past 5 years.

• Analyzed the trend and seasonality of data using statistical methods. Found event variables such as petrol price, Consumer Price Index, Calendar events such as Marriage, Harvest, and Festive has major impact on sales over the years.

• Bulit an SARIMAX model to predict the next month sales which resulted in 75% accuracy. To improve accuracy, built an Xgboost regression model considering all event variables as categorical values which resulted an accuracy of 90%.

• To predict monthly sales for different variants of two wheelers, used the similar Xgboost model which resulted good accuracy for some variants. Hence developed another confidence model which verifies whether this algorithm gives good predictions on a particular variant.

• Further increased granularity to predict sales at the dealer level for each of the two-wheeler variants with an accuracy of 87%.

Customer Feedback ticket Valid or Invalid Classification

Created an entire end to end automation process using power automate from receiving feedback ticket from customer to assigning to the respective team and stored the data in share point as list items.
When a new item is added in share point, the SVM algorithm used for valid or invalid classification triggers and assigns value to that list.
If the ticket is valid, notification will be sent to respective team else the item gets discarded.
This project achieved 98% efficiency and cost savings of $30k. The SVM algorithm resulted an accuracy of 88%.
Tools Used: Python - pandas, pickle, SVM, seaborn, Numpy.