Bhaskar Turkar

Developed QUIN: a tool for extracting insights from tabular data using large language models. Facilitated natural language interaction with data, providing textual insights and visualizations. Utilized GPT-3.5 Turbo and GPT-4 for generating SQL queries and Python code. Performed prompt engineering to optimize model performance and ensure accurate outcomes. Integrated with various databases including Google BigQuery and Azure SQL for real-time insights. Created a Streamlit front-end interface, enhancing user experience with personal datasets. Executed extensive testing for accuracy, achieving 97% in single and 92% in multi-table tests. Leveraged RAG and Langchain SQL agents for this Tool. Authored a white paper on the exploration and testing of different models and methodologies. Developed a custom image retrieval tool capable of retrieving relevant images from a database based on user queries and uploaded images. Utilized Azure AI Vision Vectorize Image and Text API to generate embeddings for user queries and stored images, enabling efficient comparison. First, leveraged Azure AI Search services for similarity searches, using images embeddings to identify images relevant to user queries and uploaded images. Employed the GPT-4 model for secondary filtering, utilizing descriptions of the initially filtered images alongside user queries to refine image selection. Effectively presented the relevant images through a user interface, achieving a commendable accuracy rate exceeding 95%. Engineered a Sentiment Analytics Tool for analyzing social media game reviews, leveraging Python and the OpenAI API. Performed data preprocessing on Game review dataset to ensure analysis readiness. Finetuned the prompt to accurately extract client-specified entities from reviews using LLMs. Crafted a Streamlit-based interface, prioritizing ease of use for diverse end-users.