GSoC 2026: Declarative API for Bayesian Survival Analysis

Hi Chris, Bill (Mentors) and the PyMC community,

My name is Mithilesh, and I am putting together a proposal for the GSoC 2026 Bayesian Survival Models project.

I really love the idea of building a high-level declarative interface (analogous to CausalPy) to hide the tedious graph construction for time-to-event models. To test the waters, I built a lightweight Proof of Concept notebook that abstracts the PyTensor math for right-censoring into a clean .fit() API.

My proposed architecture focuses on:

  1. A PyTensor abstraction layer for right, left, and interval censoring using stable operations (e.g., ).

  2. Base parametric classes (Weibull, Log-Normal) and semi-parametric classes (Cox PH).

  3. A user-friendly formula API.

  4. Stretch Goals: Adding hierarchical frailty models and basic posterior predictive checks (C-index/Kaplan-Meier overlays).

I would really appreciate any feedback on the PoC or the timeline in my proposal draft. Specifically, I’d love to know if there are any specific PyTensor edge-cases regarding interval censoring that you’d like me to address early on.

Looking forward to contributing!

Best,

Mithilesh