📣 Disclaimer

This tool is free for everyone but the costs of API calls to the models are on the user. With gemini flash and flash lite, I estimated a cost of 0.001€ per question, around 1€/1000 questions (which is a lot of questions).

---

AI Setup Engineer

AI Setup Engineer is a tool which aims to enhance the Assetto Corsa Competizione experience for players who are not akin to setups or players who just want a prototype setup to start with, and spend less time tweaking settings.

How it works

AI Setup Engineer uses artifical intelligence to guide you setup your car by exploiting domain knowledge, provided in the documents under the docs folder. You can edit them to add, remove, edit information as you see fit, but remember they will be the source of the answers you seek.

AI Setup Engineer will NOT give you exact numbers, for example it will not tell you set your camber to -2.5. However, it will give you directions to follow, for example increase your rear wing. Given this, the level of detail of your inputs is crucial to obtain tailored suggestions: for example, avoid generic questions such as reduce oversteer and prefer something like reduce oversteer in turn 2 at the exit of the corner.

Key Features:

• Identifies corners from your questions;
• Can use JSON setups as a starting point for outputting tailored suggestions;
• Uses an LLM-workflow (not a single prompt) to process your questions and get the most context out of the selected track and car, potentially avoiding too generic suggestions;
• Works with Google and OpenAI models;
• Allows restoring past chats by logging in with Discord (this is a bit overkill for personal use but can be useful if AI Setup Engineer is hosted in a server for using it with friends/fellow racers);
• All your data stays local (API calls are usually not used by providers to improve their models, and DB is local).

Project Structure

The project is organized as follows:

• data/: Contains the metadata files used by the application, such as cars and tracks data, and utilities;
• db/: Contains embeddings from the original files in docs and an init script to initialize the database;
• customization/: Contains user-configurable scripts to customize the app.
• docs/: Contains the original documents from which knowledge is extracted;
• logs/: Once the application is started, logs will be stored and persisted here;
• envs/: Contains the .env file which stores credentials and settings;
• src/: Contains the source code of the application.

Customization

Please start your scripts from the main folder so that logs and imports are managed correctly.

Customization Folder

The customization/ folder allows users to modify the behavior of the project without directly altering the core code. Place custom scripts or configuration files here.

Example: If you want to regenerate embeddings for your documents (suppose you made changes to them), use the customize_embeddings.py script and your db will be updated on the next restart.

Environment Variables

The following environment variables can be used to configure the project:

• OPENAI_API_KEY: [OPTIONAL] Your OpenAI API key (if you want to use OpenAI's models);
• OPENAI_ORG_ID: [OPTIONAL] Your OpenAI organization ID (if you want to use OpenAI's models);
• OPENAI_PROJECT_ID: [OPTIONAL] Your OpenAI project ID (if you want to use OpenAI's models);
• ENVIRONMENT: Any value you want, e.g. development, staging, production. Remove it to test locally via python-dotenv;
• MONGODB_URI: Self-explanatory;
• MONGODB_INDEX_NAME: The index name fot the embeddings;
• MONGODB_DOCS_DB_NAME: Self-explanatory;
• MONGODB_DOCS_COLLECTION_NAME: Self-explanatory;
• MONGODB_SESSIONS_DB_NAME: Self-explanatory;
• MONGODB_SESSIONS_COLLECTION_NAME: Self-explanatory;
• BASE_URI: URL of the app, e.g. http://localhost:8080;
• DISCORD_CLIENT_ID: [OPTIONAL] Set-up at https://discord.com/developers/applications;
• DISCORD_CLIENT_SECRET: [OPTIONAL] Set-up at https://discord.com/developers/applications;
• USE_VERTEXAI: [OPTIONAL] Any value to use Gemini models via VertexAI;
• GOOGLE_PROJECT_ID: [OPTIONAL] Your Google Cloud Project ID, if using VertexAI;
• GOOGLE_APPLICATION_CREDENTIALS: [OPTIONAL] Path to your Google Cloud service account credentials file, if using VertexAI. Can be mounted as a volume in Docker by setting it to /credentials/svc.json and naming your JSON accordingly;
• GOOGLE_PROJECT_REGION: [OPTIONAL] Your Google Cloud Project region, if using VertexAI;
• SUPPORT_LLM: The LLM to use for support tasks;
• RECOMMENDER_LLM: The LLM to use to generate recommendations;
• EMBEDDINGS_MODEL: The model to use for generating embeddings;
• EMBEDDINGS_DIMENSIONS: The number of dimensions of the embeddings model;
• CACHE_HIT_DESIRED_RATIO: The desired ratio of cache hits when the same question has been asked in the past. Lower values can increase diversity, higher values can lead to lower costs;
• LOG_LEVEL: The desired log level for the whole application.

Getting Started

Prerequisites

• A terminal (if you are using Windows, Docker should work but for maximum compatibility use WSL);
• Git;
• Docker;
• [OPTIONAL] A way to query the underlying MongoDB database, if needed (e.g. VS Code MongoDB extension)

Installation

1. Clone the repository:

   git clone https://github.com/andreasntr/open-ai-setup-engineer
   cd open-ai-setup-engineer # cd is dir in Windows

2. Fill the envs/.env file with the envs you need and add Google credentials, if using VertexAI;

3. Start the app:

   • Automatically (using the provided start.sh script):

     sh start.sh

   • Manually:

     • docker compose build
     • docker compose up -d

Accessing the Application

Open your web browser and go to http://localhost:8080.

License

This project is licensed under the MIT License.

A note from the developer

This project has been developed with passion during my spare time over the course of something less than a year. I reckon something is not perfect but I also reckon it would be impossible for me to keep up with the development alone.

I truly believe in open-source software and thus I believe this project can become extremely better with the contribution of any of you fellow developers-racers.

If you want to contribute, feel free to open an issue or a PR (better).

Thank you and enjoy your AI Setup Engineer.