📍 London, UK | 🤖 Lead Backend Engineer | 🚀 darlington.dev
First neural network in 2017. 8 years of backend engineering across regulated industries. Now building autonomous AI agent infrastructure. Targeting 10,000 iterations this year — every decision is training data.
graph LR
subgraph Input
Agent1["Agent"]
CLI["CLI"]
API["Python / Rust SDK"]
end
subgraph Cortex
Store["Node Store"]
Embed["Embedding Index (HNSW)"]
Graph["Knowledge Graph"]
Decay["Decay Engine"]
Briefing["Briefing Generator"]
end
subgraph Output
Context["Agent Context"]
Search["Hybrid Search"]
Prompts["Prompt Selection"]
end
Input --> Store
Store --> Embed --> Graph
Graph --> Decay
Graph --> Briefing --> Context
Graph --> Search
Store --> Prompts
style Store fill:#f59e0b,stroke:#d97706,color:#000
style Graph fill:#8b5cf6,stroke:#7c3aed,color:#fff
style Briefing fill:#10b981,stroke:#059669,color:#fff
Cortex — embedded graph memory for AI agents. Agents store knowledge as typed nodes → auto-linking via embeddings → decay of unused knowledge → briefings synthesised on demand → agents get smarter. One binary. One file. Zero dependencies. Currently trialled by 2 startups.
| Service | Purpose | Repo |
|---|---|---|
| Warren | Agent lifecycle manager (Docker Swarm) | Go |
| Dispatch | Task broker — 11-factor scoring, NATS delivery | Go |
| Alexandria | Knowledge layer — secrets, semantic search, context | Go |
| PromptForge | Prompt registry — versioning, branching, subscriptions | Python |
| Chronicle | Observability — transcript storage, DLQ alerts | Go |
| MissionControl | 10-stage orchestration — gates, workers, audit trail | Go + Rust |
| Dredd | The Judge — extracts decisions, builds trust scores | Go |
| CC Sidecar | Watches Claude Code sessions, publishes to NATS | Go |
| Agent | Role |
|---|---|
| 🦊 Kai | King. Orchestrates work, thinks architecturally. Always-on, bare metal. |
| 🌸 Lily | PA. Conversational, WhatsApp + Slack. |
| 🔭 Scout | Research specialist. Web search, structured reports. |
| ⚒️ Celebrimbor | PromptArchitect. Designs and evolves agent personas. |
| ⚔️ DutyBound | Developer. Spawned on-demand for code tasks. |
| ⚖️ Dredd | Judge. Extracts decisions from transcripts, feeds trust loop. |
Darlington — Personal OS. Habits, health, finance, Mandarin, calendar, Kai chat. Next.js 15 + Supabase.
OpenGlass — iOS smart glasses app connecting Meta Ray-Bans to Gemini Live + OpenClaw.
Cortex — Embedded Rust knowledge graph for agent memory.
- Architecting multi-agent workflows with MCP at Cox Automotive
- Building Cortex — embedded Rust knowledge graph for agent memory, trialled by 2 startups
- Contributing to OpenClaw
- Targeting 10,000 iterations this year
- Lead Backend Engineer at Cox Automotive
- Previously BNY Mellon, Finova (Bain Capital portfolio)
- First Class Honours, Computer Science — AI specialisation
- Cortex trialled by 2 startups
- Deployed LLM-powered chatbot at BNY Mellon in 2021, pre-ChatGPT
10,000 iterations, not 10,000 hours. Every decision is training data. Every conversation teaches the swarm.
Random facts
- Built my first neural network in 2017 during university — a CNN for image classification
- I speak conversational Mandarin and track progress daily on Darlington
- Ran a cricket data analytics side project with Monte Carlo simulations
- I play Old School RuneScape — efficiency-focused, naturally
- My agents have a trust scoring system — they earn autonomy over time
- The name "MikeSquared" comes from a joke about having two Mikes on a team