Kartik Godawat

A Jupyter extension that I developed that converts English queries into Python code, enabling users to execute and create data analysis dashboards. The tool utilizes natural language processing (NLP) techniques such as intent classification, named-entity recognition (NER), and templating to accurately convert user queries into relevant Python code. This tool was created before ChatGPT landed and allows users to interact with data using plain English queries.

This project leverages state-of-the-art large language models (LLMs) like LangChain, ChatGPT, LLaMA, StarCoder, Code Llama, and Whisper, along with browser integration capabilities similar to Chrome. Its primary goal is to develop versatile AI Agents capable of performing a wide range of tasks, including browser navigation, user data input, and content summarization from web pages.

For example, a user can simply voice a request into their microphone, such as, "I'm unsure which TVs to buy; find me 5 TVs under budget X dollars." The AI Agent then springs into action, scanning multiple websites, processing reviews, summarizing features, and ultimately compiling a detailed report outlining the pros and cons of various TV options. This assists users in making well-informed decisions.

Given the generic nature of these AI Agents, their applicability extends beyond a single use case. They can be seamlessly deployed in various contexts such as shopping, itinerary planning, gift selection, and more, offering a versatile solution for an array of tasks.

A fun demo project that I developed using the Transformers library, where a movie poster serves as input to generate a potential plot. The project involved fine-tuning an image-to-text model to achieve this functionality.

The primary objective of this project was to shed light on the stereotypes and biases prevalent in the content we consume. By analyzing various elements such as objects, colors, and other biases present in the input image, the model generates a plausible plot as its output.