GSOC IDEA DISCUSSION 2026

[SachinDurairaj06]

Hi everyone,

I’m interested in proposing a GSoC project around improving the Turing.jl → InferenceData export workflow in ArviZ.jl so that datasets produced from Julia sampling are more consistently ready for downstream diagnostics and model-checking workflows.

While working previously on ArviZ.jl documentation (citation fixes) and also contributing to the conda-forge packaging discussion around the modular arviz-base / arviz-stats / arviz-plots structure, I noticed that several ecosystem boundaries depend heavily on how sampler metadata and group structure are exported into InferenceData. In particular, some diagnostic-related metadata from Turing outputs is not always mapped in a way that makes the resulting objects immediately usable in typical ArviZ analysis workflows.

A direction I am considering for GSoC is:

• auditing the current Turing.jl → InferenceData conversion path
• improving export of sampler diagnostics into sample_stats where appropriate
• adding validation checks for exported InferenceData objects
• writing parity tests for a representative Julia Bayesian workflow
• documenting a simple end-to-end sampling → diagnostics example

Since this touches both ArviZ.jl export behavior and expectations shared with downstream ArviZ diagnostics workflows, I would really appreciate feedback on whether this fits current priorities and where the biggest gaps currently are.

Tagging @sethaxen , @OriolAbril , and @aloctavodia for suggestions on whether this direction would make sense as a GSoC project and which sampler fields or workflows would be most useful to focus on first.

[SachinDurairaj06]

Sorry for mentioning again @sethaxen , @OriolAbril and @aloctavodia may i receive your suggestion at the earliest

[sethaxen]

Hi @SachinDurairaj06, thanks for the proposal. Unfortunately, we currently only have one developer (me! 👋) maintaining the Julia packages, and I don't have the availability to mentor a student this year (hence why we haven't listed any open ArviZ Julia projects). If any of the Python ArviZ projects are of interest to you, you could try submitting a proposal there. Alternatively, you could submit a proposal for a Turing.jl project: https://turinglang.org/news/posts/2026-02-21-gsoc/