#uncertainty-quantification

Trustworthy AI for satellite conjunction triage using calibrated ML, Bayesian logistic regression, and uncertainty-aware escalation on public CDM data.

RAG over industrial maintenance manuals that abstains with a calibrated error guarantee: conformal risk control for selective QA, a from-scratch agent, and a prompt-injection suite enforced in CI.

Token-level entropy trajectories from LLM logprobs. Models can measure their own uncertainty — grounding it in truth requires external verification.

Backend-agnostic black-box hallucination & RAG-faithfulness detection for LLMs — Probabilistic Certainty & Consistency (arXiv:2601.02574). Works on MLX / OpenAI / vLLM via token logprobs; no model internals, no training.