Treasure Hunt Admin Bot, written on top of Langchain and GPT 3.5, for a Treasure Hunt that I organised. See this notebook to see the bot in action, including the prompts used to direct it.
Frontend (with screenshots) lives here
- Python 3.9
- Langchain and GPT 3.5 (optimised for chat)
- DynamoDB for chat history
- Serverless Framework + AWS Lambda/API Gateway
You'll need to set up the environment variable OPENAI_API_KEY with your own OpenAI key. From there on, spin up the notebook using jupyter notebook or using Visual Studio's Notebook plugin.
Instead of relying on the bot to remember which state (treasure hunt riddle) it should be on, it is instead forced to simulate a state machine. In each state, it should figure out the best action to take. The rules of the state machine are explained in simple natural language, as opposed to a formal representation.
(TODO: add screenshot for the state machine governing the bot)