GSoC 2026 Interest: Modernizing Brian2 Serialization with HDF5 and Apache Fury

Development
1

Hi Marcel and the Brian2 Team,

I am Siddhant, a CSE student and open-source developer. I am writing to express my strong interest in the Serialization/Deserialization for Brian models project for GSoC 2026.

After researching the current pickle-based store/restore mechanism in network.py, I’ve developed a proposal for a Tiered Serialization Suite to address two distinct needs:

  1. Scientific Archival (HDF5): A hierarchical container using h5py to store model architecture (JSON) and numerical state variables together. This ensures long-term reproducibility and cross-platform compatibility (MATLAB/R).

  2. Performance Checkpointing (Apache Fury): Utilizing the Apache Fury framework for ultra-fast, multi-language binary serialization to replace the runtime bottlenecks caused by pickle in large-scale simulations.

I have already developed a functional Proof of Concept (PoC) for the HDF5 archive structure and would love to share my initial findings. Does this dual-pillar approach (Archival vs. Performance) align with your vision for the project?

Best regards,

Siddhant