I design and deploy multi-agent AI systems that run businesses

Not chatbots. Not automations. Full autonomous agent architectures — orchestrated, monitored, and production-hardened.

30+ Autonomous Agents
15+ Production Clients
24/7 Running in Production
// the system

Architecture, not magic

A single orchestrator manages 30+ specialized AI agents. Each agent has its own context, memory scope, and domain expertise. They share a knowledge graph, monitor each other, and recover from failures automatically.

┌─────────────────────────────────────────────────────┐ │ BigBoss // orchestrator │ │ ├── delegates tasks to agents │ │ ├── monitors health & restarts failures │ │ └── runs nightly audits & maintenance │ ├─────────────────────────────────────────────────────┤ │ Agent Fleet // 30+ specialized agents │ │ ├── client agents (ads, seo, reporting) │ │ ├── infrastructure (sync, monitoring, dashboards) │ │ └── each: own tmux session, own memory scope │ ├─────────────────────────────────────────────────────┤ │ Shared Memory // knowledge graph (agent-recall) │ │ ├── entities, relations, observations │ │ ├── scoped visibility (client isolation) │ │ └── MCP protocol for agent access │ ├─────────────────────────────────────────────────────┤ │ Human-in-the-Loop // draft system │ │ ├── agents propose, humans approve │ │ ├── memory writes require review │ │ └── dashboard for fleet management │ └─────────────────────────────────────────────────────┘
// proof of work

Built and running, not pitched

open source

agent-recall

Persistent memory for AI coding agents. Shared knowledge graph with scoped visibility, MCP-native, 321 tests. Extracted from production into a standalone Python package.

github.com/mnardit/agent-recall →
production

30+ Agent Fleet

Autonomous AI agents managing 15+ real clients daily. Google Ads, SEO, reporting, WordPress, DevOps — each agent specialized, all orchestrated by a single system.

product

AI Audit Framework

Productized AI-readiness audit. Automated crawling, 5 parallel analysis agents, deterministic report generation. No hallucinations in client-facing output.

// about

Max Nardit

Data & AI Engineer. 10+ years turning data into working systems. Currently building the infrastructure layer that AI agents need — memory, orchestration, and human oversight. Head of Data Analytics by day, multi-agent architect by nature.

max.nardit.com →

Let's build something

Interested in AI agent architecture for your business? Let's talk.

[email protected] GitHub LinkedIn