Show HN: Viveka: filter LLM output against a Lean-verified Advaita Vedanta model (github.com) AI

Show HN introduces Viveka, a “witness-centered filter layer” for LLM apps that sits between a model and its user, extracts claims/postures from responses, and checks them against Lean-verified axioms from Scherf Logic’s formal Advaita Vedanta to flag or block objectifying, steering, or overconfident framing. The project is offered as the PyPI package “witness-layer,” with a strongly recommended [scherf] extra for the verified backend and options to FLAG/CORRECT/BLOCK rather than silently rewriting output, emphasizing a clear “honesty boundary” where claim interpretation is heuristic while the underlying axiom checks are machine-verified.