autonomous-mall-assistant is an AI-powered mall assistant designed to help shoppers easily locate stores within a shopping mall. The system utilizes a Large Language Model (LLM) to understand user queries and provide precise information or alternative suggestions.

1. User interacts with the chat interface to ask about a specific store or retail category.
2. The LLM identifies the store name and retail category from the user's query.
3. The system searches the database (pandas df) for the store information.
   • If found, returns relevant details to the LLM as context data.
   • If not found, suggests alternative stores from the same retail category and return it to the LLM to propose to the users as alternatives.

Stores Database - stored as a pandas dataframe but you should be able to use any other DBs (e.g. SQL/NoSQL) as long as you can wrap it into a function.

Note: more examples in autonomous-mall-assistant.ipynb

Mentions mall wide events, if any.

• Add mall wide events/promotions to the output - Completed on [2023-11-05]
• Add mall general info like opening hours; exceptions to the scheds because of special holidays, etc.
• Define handling of multiple stores/categories in user input.
• Incorporate some guardrail mechanism e.g., nemo guardrails.
• (longish-term) Breakdown the single search_store function into their respective retail category functions to handle targeted queries (e.g., sports - "Does Nike sell football boots?", food - "Which restaurants offer halal food?")
• (longish-term) Utilize a better database that can scale better (e.g., managed DB offerings from Azure, AWS, etc.)
• (longish-term) Add GPT-4 tool/function for complex queries requiring reasoning ability.

• GPT-4
• LangChain
• pandas
• streamlit
• streamlit-chat

1. clone this repo
2. pip install requirements.txt
3.1 run autonomous-mall-assistant.ipynb OR
3.2 cd to streamlit_frontend and run `streamlit run app.py` in terminal

...is welcome! 🤗

• [2023-11-05]: Implemented Chat History feature to maintain conversation context. Since LLM API interactions are stateless, this enhancement captures and sends the entire history of user-AI exchanges alongside the current user query, enabling the LLM to generate contextually relevant responses.

Feel free to connect with me on:

Linkedin: https://www.linkedin.com/in/stephenbonifacio/
Twitter: https://twitter.com/Stepanogil

Did you find this useful? If you'd like to show some love, I won't say no to a cup of coffee! 🤗