Introduction

This project is an implementation of question answering sql rag in InfluxDB and langgraph.

Graph

Prerequisites

• Docker installed.
• An account in Langsmith.

Models

• llama3.2
• llama3-groq-tool-use

LLM Connection

• OLLAMA

Memory

Currently, the project supports the following memory savers and switching them in .env file:

• MemorySaver
• PostgresSaver

Development

1. Clone the repository.
2. Copy the .env.example to .env and update the values as per your environment. Some value would be available after up and running the docker containers.
3. Up the docker containers

   docker-compose up -d --build
   

InfluxDB

1. Run

   docker-compose exec -it influxdb3 bash
   

2. Create the token:

   influxdb3 create token --admin
   

3. Set INFLUXDB3_AUTH_TOKEN:

   export INFLUXDB3_AUTH_TOKEN=admin_token
   

4. Create the database:

   influxdb3 create database sensors
   

5. Write into the database:

   influxdb3 write --database sensors  --file /home/data/air-sensor-data.lp
   

   influxdb3 write --database sensors  --file /home/data/sensor-meta-data.lp
   

6. Verify:

   influxdb3 query --database=sensors "SHOW TABLES"
   

7. Delete if needed:

   influxdb3 delete table --database sensors air_sensors
   

    influxdb3 delete table --database sensors sensor_meta
   

Update the .env file with the token and database name.

Ollama

1. Run

   docker-compose exec -it ollama bash
   

2. Install models:

   ollama pull llama3.2
   

   ollama pull llama3-groq-tool-use
   

3. Start the server:

   ollama serve
   

Langgraph Studio

1. Run

   docker-compose exec -it langgraph bash
   

2. Run

   langgraph dev --host 0.0.0.0 --port 2024
   

3. Studio UI. On failure, browse to Langgraph Platform > Langgraph Studio.
4. Langsmith UI.

CLI

1. Run

   docker-compose exec -it langgraph bash
   

2. Run

   python rag.py
   

3. To end the chat type exit or quit.

Frontend UI

1. Adjust views/.env
2. Go to: http://localhost:8501/ or replace localhost with the IP address of the machine.

Testing

It is recommended to perform unit test before commiting the code. To run unit test, run the following command

pytest

Type Checking and Linting

This repo uses pre-commit hooks to check type and linting before committing the code.

Install:

pip install pre-commit

Enable:

pre-commit install

License

This project is licensed under the MIT License - see the LICENSE file for details.