Ongoing & Upcoming
Ideas I may build next
- REB v1.0 / REB Expansion - workload generators, cleaner comparison plots, reproducibility checks, and hooks for real robot logs.
- GNN Revisitation - second pass at VLE models with identity failures tested upfront and classical-property fallback treated as part of the system.
- AI Responses to High-Risk Topics - study of how high-risk prompts activate internal model features, refusal behavior, and failure modes. Combines safety-preserving prompt design, response evaluation, and activation-level analysis in open-weight models.
- Scientific Surrogates With Error Bars - small-model uncertainty, calibration, and physics-aware checks for scientific surrogate decisions.
- Hardware-Forward Testbeds - possible REB-H and/or EcoBench builds for physical energy, sensing, control, and MRV-style experiments.
- Smart Carbon MRV System / Micro Digital Twins - possible longer-run work combining simple forward models, sensor streams, uncertainty estimates, and verification workflows.