A Robust Tabular Data Question Answering Engine for Electricity - **Developed an advanced question-answering (QA) engine** tailored for the electrical sector, utilizing **TAPAS and Retrieval-Augmented Generation (RAG)** techniques to handle complex tabular data queries. - **Worked extensively with tabular data**, extracting and structuring information in JSON format from **Central Electricity Authority (CEA) APIs** of Peak energy and Energy requirements of states of… Show more A Robust Tabular Data Question Answering Engine for Electricity - **Developed an advanced question-answering (QA) engine** tailored for the electrical sector, utilizing **TAPAS and Retrieval-Augmented Generation (RAG)** techniques to handle complex tabular data queries. - **Worked extensively with tabular data**, extracting and structuring information in JSON format from **Central Electricity Authority (CEA) APIs** of Peak energy and Energy requirements of states of India.- **Addressed limitations of TAPAS's SoT model** for tabular data by implementing RAG, which significantly improved context handling and retrieval accuracy.- **Innovated a novel data chunking method**, effectively overcoming TAPAS's 512 token limit, thereby expanding the model's context capacity and enabling more comprehensive queries. - **Discovered and implemented a new method for chunking and feeding data** into the retrieval model, enhancing its ability to handle larger datasets and improving overall system efficiency.- **Implemented LLM model architecture Wizard LLM2** to enhance the interpretation of natural language queries and to manage sophisticated, context-dependent questions more effectively.- **Integrated Qdrant vector database** for efficient embedding storage and retrieval, and utilized **Cohere's rerank model** for prioritizing the most relevant content by leveraging the Cosine Similarity.- **Created an interactive dashboard** using Streamlit to visualize query results and data insights, making it easier for users to interact with and understand the data.- **Conducted rigorous human verification processes** to ensure the relevance, quality, and factual correctness of the generated solutions, enhancing the reliability of the system.- **Successfully enhanced tabular data query capabilities**, significantly improving upon traditional SQL-based solutions, making the QA engine more intuitive and effective for specialized . Show less

𝗔𝗿𝘁𝗶𝗳𝗶𝗰𝗶𝗮𝗹 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝗳𝗼𝗿 𝗜𝗻𝗱𝘂𝘀𝘁𝗿𝗶𝗮𝗹 𝗔𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 :-1. Developed an Electric Vehicle range prediction model incorporating internal and external factors using LSTM and ensemble models (Random Forest).2. Created a model for Compatibility Time Prediction utilizing XGBoost.3. Implemented Driver Behavior Recognition through clustering techniques.