I build memory and security infrastructure for AI agents. Creator of Aegis Memory (persistent memory for multi-agent systems) and mcp-parapet (MCP security middleware). Founder of Quantify Labs Ltd.
This site is where I write about what I learn building these systems — agent memory architectures, context engineering, multi-agent orchestration, and the messy production details that tutorials skip.
-
Scope-Aware Memory Access Control for Multi-Agent Systems
How do you let Agent A read shared knowledge without leaking Agent B's private context? I built a 3-tier memory scope system inside Aegis Memory. Here's the architecture, the trade-offs, and what broke along the way.
-
I Tested 6 Agent Memory Types on the Same Conversation — Here's What Survived
Buffer, summary, vector, graph, entity, and token-windowed memory. Same 20-turn conversation, measured on token cost, retrieval accuracy, and context quality. The results weren't what I expected.
-
The CrewAI + LangGraph Hybrid Pattern Nobody Talks About
CrewAI handles team orchestration. LangGraph gives you explicit state control per agent. Using both together is the pattern I wish I'd known earlier. With code and architecture diagrams.
-
Context Window Management in Production: Token Budgets, Selective Retrieval, and What Actually Matters
Your agent has 4K tokens of context budget and 50K tokens of potentially relevant memory. How do you pick the right subset? Recency-only vs relevance-only vs hybrid scoring — tested on real workloads.
-
The Validation Gap in Multi-Agent Simulation: Lessons from OASIS and TRIBE
Multi-agent simulations look impressive but how do you know the emergent behaviours are meaningful and not just hallucinated consensus? A critique that became my PhD research direction.
-
Building MCP Security Middleware: Content Scanning, Trust Boundaries, and the EU AI Act
MCP gives agents access to external tools. mcp-parapet sits in between to scan content, enforce trust boundaries, verify manifest integrity, and create audit trails. 2,778 lines of Python, 63 tests, and a compliance story.
About me
I'm an AI infrastructure engineer and founder of Quantify Labs Ltd. (UK). MSc in Big Data Analytics from Sheffield Hallam University. Currently building toward an AI Agent Engineer role and a funded PhD researching multi-agent memory systems and collective knowledge discovery.
Previously: 10+ years in data engineering and BI. Azure & Databricks certified. 1,700+ LinkedIn followers writing about agent memory and AI infrastructure.