auditable-ai - 技术专题

The GEP-powered self-evolving engine for AI agents. Auditable evolution with Genes, Capsules, and Events. | evomap.ai

A proactive AI agent for secure, traceable, human-in-the-loop task execution over long-running workflows.

Auditable offline edge intelligence for low-cost edge devices, with benchmark evidence and public board proof on ESP32-C3.

An accountability layer for investment agents: thesis cards, prediction ledgers, stock screeners, and feedback loops.

I've always believed that the hardest problems in AI aren't technical; they're architectural. We keep building systems that can't explain themselves, can't prove their own integrity, can't handle uncertainty without either freezing or lying. And then we act surprised when we don't trust them.

An experimental Application Object Model for helping LLM agents understand, reason about, and operate real apps.

Toward controlled evolution of artificial intelligence through validated neural grafting.

paceval is a high-performance mathematical runtime for deterministic AI and energy-efficient edge computing.

A governance layer for autonomous AI agents, proven on a life: 30+ scheduled agents made auditable, self-healing, and human-gated by single-writer field ownership. Architecture and patterns, no personal data.

Defensive preprint, governance, and threat model for DOORM Genmount WorldModel-OS — a governance-first architecture for auditable agentic reasoning.

KnowMesh / 知络: local-first knowledge asset compiler for auditable RAG, traceable citations, and maintainable document intelligence.

Outer-project research shell for the Vibe Science kernel.

ECHO-PPI: Evidence-bundled, auditable overlapping protein-module detection in protein-protein interaction networks.

Code and lightweight audit artifacts for auditable repair of scientific reasoning graph extraction on a 350-row benchmark.

A self-distilling neuro-symbolic cascade that amortises LLM cost across knowledge-graph QA and regulatory-compliance checking, with auditable Datalog proof trees.

L1 platform skeleton for DOORM WorldModel-OS — gateway, discrete-anchor state space, append-only audit trail, and four-level rollback contract for auditable agentic reasoning. Code release pending §11 gates.

Bank Shield: Auditable AI for credit scoring that maintains accuracy while enforcing regulatory compliance (66% fewer breaches), proving resistant to adversarial text manipulation.

GATOS generalizes Git’s content-addressed DAG into a complete computational substrate: a programmable operating surface unifying history, policy, and computation within a deterministic, verifiable feedback loop.

Local-first memory and simulation infrastructure for auditable AI agents. Index evidence, build knowledge graphs, run bounded simulations, and produce reviewable verdicts on local disk.

Open-source memory and experience layer for persistent AI agents, with auditable decisions, evidence-based learning, and evolving identity.