I build AI systems that operate in the real world.

My work sits across AI systems architecture, platform engineering, DevOps, cybersecurity, connected-device operations, business-system integration, and research engineering. I design the runtime, wire the infrastructure, secure the boundaries, integrate existing systems, and turn messy operational workflows into intelligent, measurable platforms.

I specialise in private/local AI systems where sensitive data, infrastructure access, and business workflows need to stay controlled.

The interesting part is not prompting a model.
The interesting part is building the system around it — securely, observably, and in production.

• AI systems architecture: agent runtimes, RAG, model routing, MCP integrations, local LLM infrastructure, evaluation loops.
• Platform & DevOps: AWS, Linux, Docker, systemd, Cloudflare, CI/CD, observability, backups, deployment automation.
• Cybersecurity: threat modelling, secure agent boundaries, pentesting workflows, policy gates, audit trails, secret handling, access control.
• Backend systems: API design, authentication, data modelling, workflow automation, event handling, and integration architecture.
• Operational intelligence: dashboards, anomaly detection, SLA tracking, diagnostics, support automation, and account intelligence.
• Research engineering: controlled experiments, matched arms, falsifiers, evidence chains, local model labs, LoRA and state-continuity work.

Syntra  ·  adaptive decision runtime
Self-hosted decision infrastructure for context-aware policy execution, model routing, delayed-feedback learning, persistence, auditability, and operational decision loops.
Demos include LLM routing, anomaly routing, autoscaling, pandemic-policy simulation, live NASA/JPL HORIZONS Mars planning, and edge-of-chaos numerical detection.

Portreeve  ·  security layer for LLM agents
Host-side policy and safety primitives for agent platforms: capability policy, approvals, audit logs, run budgets, and session-aware lethal-trifecta detection.

Synthena  ·  public research notebook
An investigation into continuity, memory, embodiment, persistent state, mortality, and lineage in language-model-based systems. Not an AGI claim, consciousness claim, or life claim.

Stash  ·  AI coding session manager
A practical tool for resuming Claude, Codex, and OpenCode sessions across projects.

MoEfolio  ·  AI market debate experiment
A public paper-trading research desk where multiple AI personas debate stock picks, execute simulated trades, and track outcomes.

• Runtime-first. I build the control layer around AI: what it can see, do, remember, call, and affect.
• Security-aware. I treat tools, credentials, user data, model output, and automation boundaries as first-class risk surfaces.
• Operations-grounded. I build for real workflows: support teams, devices, customers, SLAs, audits, integrations, and failures.
• AI-leveraged. I use agents aggressively to move faster, while retaining ownership of architecture, tests, security, and claims.
• Evidence-disciplined. Claims need controls. Findings need artifacts. Nulls and retractions stay visible.

Open to senior, staff, principal, founding, or consulting conversations around AI systems architecture, agent infrastructure, platform engineering, secure automation, private/local AI, and operational intelligence.

• X: @ashxhart

The work above is real, current, and built with AI as a force multiplier.