BIM-IoT Assistant

Description

BIM-IoT Assistant is a chatbot that integrates Building Information Modeling (BIM) data with Internet of Things (IoT) sensor data to provide intelligent responses to queries about buildings and their sensor readings. The system uses a knowledge graph (GraphDB) to store building information and time-series database (InfluxDB) to store sensor data.

Relevant Directories

• competency-questions/: Contains a PDF with questions the system should be able to answer
• prompts/: Contains prompt templates for the LLM to generate SPARQL queries
• dataset/: Contains the data files for the smart home example

Requirements

• Docker compose
• OpenRouter API key
• Python 3.8+
• GraphDB instance
• InfluxDB instance
• Neo4J instance

Note: current implementation uses OpenRouter as the default LLM API provider. However, the system is designed to be flexible - you can easily switch to a different LLM API provider by modifying the implementation in llm.py. This modular design allows for easy integration of custom LLM solutions or alternative API providers.

Installation and Setup

This step is only required to import the initial dataset into the databases.

1. Clone this repository
2. Create and activate a virtual environment

   # Windows
   python -m venv venv
   venv\Scripts\activate
   
   # macOS/Linux
   python -m venv venv
   source venv/bin/activate
   

3. Install the required dependencies:

   pip install -r requirements.txt
   

Database Setup

You can run the databases using docker compose.

1. Copy the provided .env.example in .env and edit the file based on your preferences.

2. Start the docker compose with the command:

   docker compose up -d
   

3. You can check the logs with docker compose logs -f

Now, you should be able to connect at

• InfluxDB (accessible at http://localhost:8086)
• GraphDB (accessible at http://localhost:7200)
• Neo4j (accessible at http://localhost:7474)

Configure InfluxDB (optional if you do not use the Docker Compose):

• Access the InfluxDB UI at http://localhost:8086
• Create a new organization named "BIMIoT"
• Create a new bucket named "OpenSmartHome"
• Generate an API token and save it for later use

Configure GraphDB

• Access the GraphDB UI at http://localhost:7200
• Create a new repository named "smartHome"

Application Environment Setup

To use the application, a proper dataset must be loaded in the databases. 1. 2. Copy the file .streamlit/secrets.toml.example to .streamlit/secrets.toml).

3. Edit the variable to choose the LLM model and the database type (GraphDB or Neo4j). For example:

# LLM API Key
LLM_API_KEY = "your-openrouter-api-key"

# InfluxDB Configuration

# InfluxDB Configuration
INFLUXDB_URL = "http://localhost:8086"
INFLUXDB_TOKEN = "your-influxdb-token"
INFLUXDB_ORG = "BIMIoT"
INFLUXDB_DATASET_SOURCE = "/dataset/OSH_Measurements"

GRAPHDB_URL = "http://localhost:7200/repositories/smartHome"

# Neo4j Database Configuration
NEO4J_URI = "bolt://localhost:7687"
NEO4J_USERNAME = "neo4j user"
NEO4J_PASSWORD = "neo4j user's password"

TTL_FILE_PATH = "dataset/openSmartHome_Donkers.ttl"

# graphdb (default) or neo4j
BACKEND_TYPE = "graphdb"

Data Import

IoT sensors dataset

InfluxDB default dataset refers to the sensor data CSV files defined in INFLUXDB_DATASET_SOURCE (see .streamlit/secrets.toml.example).

To import the dataset, proceed as follows:

1. Run the import script:

   python import_sensor_data.py

2. Check the imported data (optional). From the query builder of InfluxDB you can use this query:

   from(bucket: "OpenSmartHome")
   |> range(start: 0) 
   |> first() 
   

Importing RDF dataset to GraphDB

The BIM dataset to load in GraphDB refers to the TTL_FILE_PATH file defined in INFLUXDB_DATASET_SOURCE (see .streamlit/secrets.toml.example) TTL file can be imported through the GraphDB workbench interface

Importing LPG dataset to Neo4j

1. Update the path to the TTL file TTL_FILE_PATH (see .streamlit/secrets.toml.example).

2. Run the import script:

   python import_ttl_neo4j.py

Usage of the Virtual Assistant

You can select the preferred backend by changing the value of BACKEND_TYPE (.streamlit/secrets.toml.example). The default is graphdb but you can change it in neo4j.

You can restart the docker compose of the streamlit application:

docker compose stop streamlit
docker compose rm streamlit
docker compose up -d streamlit

You can also run the prototype by running the following command in the terminal:

streamlit run bot.py

To access the Streamlit web application, open the web browser at the address http://localhost:8501/

You can now interact with the BIM-IoT Assistant. You can ask questions about the building structure, elements, sensors, and their readings.

Example queries:

• "List the rooms located on the first floor"
• "Information about a room"
• "Which is the lowest temperature registered in the kitchen?"

Authors

• Daniele Parumboiu
• Massimo Callisto De Donato
• Emanuele Laurenzi

Credits

• University of Camerino (UNICAM)
• University of Applied Sciences and Arts Northwestern Switzerland (FHNW)

References

This work is inspired by:

• Callisto De Donato, M., Laurenzi, E., & Porumboiu, D. (2025). A hybrid artificial intelligence to support information retrieval in smart buildings. Joint Proceedings of the BIR 2025 Workshops and Doctoral Consortium, 4034, 157–169. https://ceur-ws.org/Vol-4034/paper79.pdf

License

This project is licensed under the BSD-3-Clause License - see the LICENSE file for details.