keriyuu/chatgpt-driven-shopping-assistant

A product-level application that customizes online shopping links for users.

ChatGPT-Driven Shopping Assistant

This is an application based on Langchain and powered by large language models (LLMs). It integrates state-of-the-art prompting tricks and allows LLMs to freely call external APIs, ultimately providing users with online shopping links.

Features

• Theory of Mind: LLMs strive to infer users' intentions as accurately as possible, enabling the handling of ambiguous user responses.
• Smart Retrieval of Tools: Utilizes a FAISS-based retriever to select relevant tools based on the similarity between a user's query and the tools' descriptions.
• ReAct Prompting: LLMs are employed to generate both reasoning traces and task-specific actions (Thought/ Observation/ Action) in an interleaved fashion.
• In-Context Memory: LLMs maintain conversation history, ensuring consistency in dialogues and facilitating some mathematical operations.
• Interactive Web Application: Designed using the Gradio library, this feature offers users a user-friendly interface to engage with the agent.
• Visualized Intermediate Steps: Allows users to observe LLMs' complete reasoning process, minimizing uncertainties regarding potential hallucinations.

Setup

1. Clone this repository to your local machine.
2. Obtain API keys for OpenAI and RapidAPI. Add these keys to a configuration file named keys.txt in the root directory of the project.
3. Install the required Python libraries by running pip install -r requirements.txt in your terminal.
4. Execute python app.py to launch the web application.

References

S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y. Cao, "ReAct: Synergizing Reasoning and Acting in Language Models,"