This project was made for the 12th edition of INTER IIT which took place in IIT Madras. The project solves the high prep problem statement which comprised of building an AI chatbot which could generate a complete JSON formatted API call code given a set of tools (could be dynamically added/removed/modified) which apon querying the with a specific question outputs nothing but the fromatted API call code.
The problem required us to utilise the COT based reasoning of LLMs, which involves breaking the complex question into sequence of simpler questions and navigating through the sequence path was inspired by the DFSDT algorithm present in the ToolLLM paper whose link will be provided as follows:
paper link : https://arxiv.org/abs/2307.16789
Our complete work including data used, data created, list of tools created, alternating approaches and their codes could be found in the following drive link provided below:
drive link : https://drive.google.com/drive/folders/1_ydX-CZQW1OrXtvR7lSp8E2wcaRe-UFS?usp=sharing
The final code file has been aploaded to the github.