I build infrastructure and the software that runs on it — AWS, Terraform, Kubernetes, and a lot of Rust. The thread through everything below: I like systems that run themselves, watch themselves, and get better without me babysitting them.

I also take this on as client work → 🌐 cloud-lord.com

Open-source contributor @ Djed Alliance

Also me on GitHub: clawddar (AI-agent experiments) and shaddar-aw.

The big one. I speak or type a task, and a fleet of AI agents takes it from there: one plans the work, others route it to the right specialist, execute it on my own infrastructure, and record what they learned so the next run is smarter. Under the hood: a Rust + Kafka task scheduler, a tiered knowledge and retrieval layer the agents share, retrieval-quality evals to catch the pipeline lying to itself, and Terraform provisioning every resource it touches — the whole platform rebuilds from scratch. It's private while I pull the operator-specific bits out; the reusable pieces are already split into standalone repos, and they'll go public one by one — the first is below.

The first AITA piece that's public — the rest are following. A small Rust service that runs Terraform on behalf of AI agents: the agent POSTs a directory, gets back a run id, and that's all it ever sees. The full plan/apply log streams to disk instead of into the model's context window, and credentials never touch the agent at all. If you run coding agents against real infrastructure, this is the difference between burning thousands of tokens per apply and burning a few dozen. Ships with an MCP door so your agents can plug straight in.

A game-stats platform, built end to end — match data flows through a Rust pipeline into a serving database, and the compute that crunches it only exists while there's work to do. The moving parts:

• Ingestion & analytics — Rust services coordinated over a Kafka event bus, with backfill and analytics as separate crates so each scales on its own.
• Dynamic infrastructure — a Kafka-command-driven provisioner spins up cloud servers when analytics or backfill work lands and tears them down when the queue drains. Every box is tracked in Postgres; nothing sits idle burning money.
• Promotion gate — processed data reaches the serving database only through a watermark-gated, alignment-checked append, so a bad batch can't make it in front of users.
• Privacy by construction — accounts become irreversible HMAC tokens before anything is stored, and an opt-out suppression gate is wired into the pipeline itself.
• Observability as code — Grafana dashboards live in Git and auto-sync to the ops stack.
• A React/TypeScript frontend on top, plus small supporting services like social-preview card rendering.

My consulting practice — AWS platform engineering, migrations, and secure LLM/agent enablement.

_{The stats only count public repos — most of the work above is still private and coming out gradually.}