BIDS-Eye: A Freetxt LLM-Powered BIDS Search Engine #15

Title

BIDS-Eye: A Natural language Search Engine for BIDS datasets

Short description and the goals for the OHBM BrainHack

Instead of painstakingly searching for BIDS datasets with certain properties, or writing scripts, you can query the datasets on www.openneuro.org/ by writing freetext questions such as: “show me datasets with resting-state fMRI and participants with Parkinson’s disease” and get structured results back.

Currently, this is operating as a proof-of-concept demonstrator on openneuro data. The endgoal is to extend the functionality to further public open-source datasets, and to allow researchers to index and query their own local data by using local models.

The demonstrator can be found on www.bidseye.xyz. For the password, please send me a private mail. This is done to prevent waste of tokens.

Brainhack Goals:

1. More Data: Extend the database schema (BIDS-SQL) to allow for more flexibility in queries and better metadata structuring.
2. Better Results: Improve the Text-To-SQL Pipeline for better query results.
3. Go Local: Explore the architecture to adapt BIDS-Eye for local collections and future local-model (LLM/RAG) deployments. Also improve the input pipeline for new, foreign data.

Link to the Project

https://github.com/paul188/BIDS-Eye

Image/Logo for the OHBM brainhack website

https://drive.usercontent.google.com/download?id=18974OAiRlZsyjECThJD0U-cZxYDCuoDm&export=download

Project lead

Paul Johannssen
@github: paul188
@discord: paul1887113
Affiliation: University Clinic Bonn, Germany

Main Hub

Bordeaux

Link to the Project pitch

www.bidseye.xyz (please contact the project lead to gain access)

Other hubs covered by the leaders

• Bordeaux
• Hybrid (Asia / Pacific)
• Hybrid (Europe / Middle East / Africa)
• Hybrid (Americas)

Skills

• Python: Intermediate (Essential for backend, indexing, and query logic)
• BIDS Specification: Solid understanding of standard BIDS structure and metadata
• SQL & Databases: Basic to Intermediate (For optimizing the SQLite backend)
• Frontend/Full-Stack: Optional (Experience to help us UI-ify and improve the search interface)
• LLM / RAG: Optional (Experience with local models to improve term resolution and SQL generation)
• C++: Optional (For optimizing heavy queries)

Recommended tutorials for new contributors

• Introduction to the BIDS Specification – Essential for understanding the structure and metadata rules of the Brain Imaging Data Structure.
• The PyBIDS Tutorial – Great for grasping how Python currently interacts with BIDS, and understanding the limitations we are trying to overcome.
• SQLite Tutorial (Basic SQL Queries) – A quick primer on SQL, perfect for working with our backend metadata database.
• FastAPI Tutorial – Learn the basics of the high-performance backend framework powering our query logic and API.

Good first issues

No response

Twitter summary

No response

Short name for the Discord chat channel (~15 chars)

bids-eye

Please read and follow the OHBM Code of Conduct

• I agree to follow the OHBM Code of Conduct during the hackathon