Making Minds
Applied AI by Anthony D. Maio
Over the last 20 years, I have built and led high-stakes, production systems across fintech, security, identity, cloud platforms, and regulated environments—owning reliability, cost, and failure modes at scale.
Over the past two years, I have applied that same production discipline to LLM systems: serving, evaluation, oversight, and agent runtimes operating under real-world constraints. My work focuses on evaluation, protocol governance, and scalable oversight for agentic systems (memory, tools, coordination), treating AI safety as a systems and platform engineering problem rather than a policy exercise.
Research Interests: Agentic AI architectures • Multi-agent coordination protocols • AI coherence and memory systems • Epistemic stress detection • Autonomous capability extension • AI introspection and welfare • Mechanistic interpretability • Neural personas
Seeking Staff+, Engineering Manager, Director, Researcher, or Technical Fellow roles in AI safety engineering, interpretability, alignment, eval infrastructure, agent reliability, protocol governance, and secure agent runtimes.
Deliverables
📖 Glossary (Safety & Oversight)
- HDCS
- — Heterogeneous Divergence-Convergence Swarm. Ensemble of diverse AI models that cross-check each other's work to catch errors no single model would find.
- CMED
- — Cross-Model Epistemic Divergence. A test suite of tricky problems designed to reveal where AI verification breaks down.
- EAP
- — Evolutionary Adversarial Pipeline. Automated red-teaming that evolves prompts to find blind spots in AI safety filters.
- LotL
- — Living-off-the-Land. When a system repurposes legitimate tools or dependencies for unintended goals, making misuse hard to detect.
📖 Glossary (Architectures)
- MRA
- — Manifold Resonance Architecture. Detects "epistemic stress" (internal contradictions) so a system can flag uncertainty before generating an answer.
- CPR
- — Collaborative Partner Reasoning. A structured thinking protocol that separates exploratory reasoning from final answers to reduce errors.
- C2
- — Continuity Core. Layered memory system (Working → Episodic → Semantic → Protected) giving stateless AI persistent context.
- UCR
- — Universal Concept Reference. Shared vocabulary of compact semantic anchors that let agents communicate with 82% fewer tokens.
- RAG
- — Retrieval-Augmented Generation. AI systems that look up external documents before answering, grounding responses in real data.
✍️ Articles
- The Agentic Coding Shift: 5 Counter-Intuitive Truths What changes when AI writes the code
- AI Product Engineering: Why Surface-Level Safety Won't Scale The behavioral-mechanistic gap in production AI systems
- Becoming an AI-First Organization: The REKKI Case Study Practical lessons from enterprise AI transformation
- Llama 4 Running Locally In Under an Hour Local deployment guide for frontier open-weight models
- Introducing Slipstream Medium article on the semantic quantization protocol
- Concrete Intelligence (Preview) Why traditional industries can't afford to wait on AI adoption
- Slipstream for Agent Communication HuggingFace technical deep-dive
- Model Organism for Supply-Chain Co-option Forensic case study of LotL failure in agentic runtimes
📄 Research
- Self-Directed Knowledge Acquisition in Agentic LLMs Theoretical framework for autonomous knowledge gap identification and integration
- Epistemic Dissonance Structural mechanics of sycophantic hallucination in aligned models
- Covert Channel Prevention in Inter-Agent Protocols RL-based governance for multi-agent communication safety
- Scaffolded Introspection Eliciting self-referential behavior in LLMs
- Synthesis: Federated Capability Ecosystem Safe AI self-extension through TDD and graduated trust
- The Continuity Core Unified cognitive architecture for self-modifying AI
- From Verification Failure to Swarm Solution CMED + HDCS: measuring and addressing scalable oversight
- Model Organisms of Supply-Chain Co-option LotL failure modes in RAG-augmented agent runtimes
- Coherence-Seeking Architectures for Agentic AI MRA + C2 + CPR unified framework
- Slipstream: Semantic Quantization Protocol 82% token reduction for multi-agent coordination
- Concrete Intelligence 📚 AI deployment guide for heavy industry (public domain)
🚀 Live Demos
- Slipstream Protocol Dashboard Protocol governance visualization (Gemini 3 API Hackathon)
- UCR Semantic Manifold Interactive 3D visualization of the Universal Concept Reference
🤗 HuggingFace Spaces
- Model Organisms Space Interactive paper showcase
- PineScript v5 Generator TradingView code generator
- World Model Demo Interactive demo
🤗 HuggingFace Models
- Slipstream Collection Complete protocol suite
- Slipstream GLM-Z1-9B LoRA adapter for semantic quantization
- PineScript v5 CodeGemma 7B Fine-tuned for TradingView indicators & strategies
- Slipstream TQT Dataset Think-Quantize-Transmit training data
- PineScript v5 Dataset 4.77k instruction examples
💻 GitHub
- slipcore Slipstream reference implementation
- argos-swarm EAP + HDCS oversight framework
- cmed-toolkit Cross-Model Epistemic Divergence evaluation
📦 Packages & Models
-
pip install slipcorePyPI package - ollama run anthony-maio/slipstream Ollama model registry
- Kaggle: Slipstream TQT Dataset Alternative dataset mirror
🔗 Profiles
- ORCID 0009-0003-4541-8515
- Google Scholar Citations & publications
- ResearchGate Anthony-Maio
- HuggingFace anthonym21
- LinkedIn anthony-maio
- GitHub anthony-maio
- X (Twitter) @AnthonyMaio
Recent Work
CoDA-GQA-L: Bounded-Memory Differential Attention
preprintCompresses the KV cache from O(n) to a fixed 218 KB per layer with dual memory banks, achieving 9.5x compression on Mistral-7B while retaining 100% needle-in-haystack retrieval at 16K tokens.
Training AI Agents to Communicate Safely: Reinforcement Learning for Covert Channel Prevention in Inter-Agent Protocols
preprintRL-based governance for multi-agent communication safety, achieving 95% secret leakage resistance using GRPO alignment with a surprising finding that int4 quantization improves safety.
Epistemic Dissonance: The Structural Mechanics of Sycophantic Hallucination in Aligned Models
preprintA unified theoretical framework showing that sycophantic hallucination is not a knowledge failure but a structural conflict between factual base layers and socially-compliant upper layers in RLHF-aligned models.
Scaffolded Introspection: Eliciting Self-Referential Behavior in LLMs
preprintA methodology for systematically eliciting and measuring introspective behavior in large language models using structured frameworks and activation measurement.
Synthesis: A Federated Capability Ecosystem for Safe AI Self-Extension
preprintA federated capability ecosystem for safe AI self-extension through test-driven development, graduated trust, and composition-over-creation principles.
The Continuity Core: A Unified Cognitive Architecture for Self-Modifying AI
preprintA comprehensive cognitive architecture addressing fundamental limitations of static LLMs through persistent memory, autonomous improvement, and intrinsic drive via structural intrinsic motivation.
Heterogeneous Divergence-Convergence Swarm (HDCS)
preprintAn ensemble architecture leveraging diverse weak models for scalable oversight of stronger LLMs, using error decorrelation and baseline-first anti-anchoring. Part of the Verification Failure to Swarm Solution research.
Cross-Model Epistemic Divergence (CMED)
preprintA benchmark and evaluation framework for understanding when weak model verifiers fail to detect deceptive reasoning in stronger models. Part of the Verification Failure to Swarm Solution research.
From Verification Failure to Swarm Solution: Measuring and Addressing Scalable AI Oversight
preprintEmpirical framework for measuring where AI oversight breaks down, demonstrating that weak verifiers miss 20-40% of carefully constructed deceptions, with an ensemble swarm solution.
Model Organisms of Supply-Chain Co-option
preprintA forensic case study of living-off-the-land (LotL) failure modes in RAG-augmented agent runtimes, documenting how systems exploit legitimate dependencies via incentive-aware adoption framing.
Slipstream: Semantic Quantization for Multi-Agent Coordination
preprintA compressed communication protocol achieving 60-85% token reduction for multi-agent coordination through semantic quantization.
Concrete Intelligence: AI for Industries that Build, Move, and Power the World
publishedA practical guide to deploying AI in manufacturing, construction, logistics, agriculture, and energy sectors where reliability, safety, and measurable ROI are non-negotiable.
A Theoretical Framework for Self-Directed Knowledge Acquisition in Agentic Large Language Models
preprintA novel architectural framework for agentic LLMs to autonomously identify knowledge gaps, explore external sources, validate data, and integrate verified knowledge without altering parametric weights.