WPVH AI Chat Solution

This repository hosts a retrieval QA solution for WPVH Eindhoven AI chat, developed using the Mistral LLM model via the Ollama application.

Demo: https://vimeo.com/907558591/310a7a27d4

Getting Started

1. Install Ollama: Visit Ollama.ai to install Ollama. For Windows, use the Docker version. You can find a tutorial here and the Docker image here.

2. Clone this Repository: Use git clone to download this repository.

3. Navigate to the Directory: Use cd vhubRetrievalQA to navigate to the relevant directory.

4. Create a System Prompt for the LLM Model: Run ollama create vhubAgent -f ./Modelfile inside the directory. Skip this step for Windows.

5. Run Ollama: Use docker exec -it ollama ollama run vhubAgent for Docker or docker exec -it ollama ollama run mistral for Windows.

6. Open a New Terminal Window: Navigate to the same directory using cd vhubretrievalqa.

7. Enter Your Credentials: Open keys.py and enter your username and password for the vhub delta API.

8. Run the Data Server: Execute python dataServerWithRoomsCSVNew.py.

9. Run the Retrieval QA Server: Open a new terminal window and run python ollamaWithDataCSV.py.

The above steps will launch a chat accessible at localhost:5003. Use the app to chat about data from WPVH sensors. The chat history is displayed, but the model is not aware of it.

Important Notes

• Use the clear data button to prevent overwhelming the data server. After clearing, some time will be needed to fill the dataframe with new values.
• The code uses threads, limited to 100, keeping the number of actively updating sensors around 100. If you increase this number in dataServerWithRoomsCSVNew by changing semaphore = threading.Semaphore(100), it will become more demanding. There are 191 sensors in the sensor_list.txt file. To get all of them concurrently updating, the Semaphore Object must also reach 191. However, this number creates too many concurrent threads and causes bugs in the program, such as saving wrong values and injecting empty values into the dataframe.

Deployment Options and Requirements

For deployment options and requirements, refer to this Google Document.