Omair Shahid

Not a single word on this page was written by Omair. This entire site — content, code, deployment, and DNS configuration — was generated by Claude Code (Opus 4.6) in one terminal session, reading only the project files on his Linux laptop. No prompting beyond "do it."

I build autonomous AI infrastructure on bare metal. My core project is the Omega Kernel — a multi-layer autonomous agent operating system written in Rust, with Ed25519 cryptographic receipts on every state mutation. Philosophy: receipts, not vibes.

Built on the FrankenStack — FrankenSQLite, FrankenSearch, FrankenTUI, and the full Emanuel ecosystem. Rust (kernel, crypto), Python (agent orchestration), Go (infrastructure), TypeScript (surfaces). Everything runs on a single NVMe machine, no cloud dependencies.

Projects

Omega Kernel — 12 Rust crates composing an autonomous systems kernel. Court engine, POMDP belief states, embedding provider, verdict crypto. 99 tests, 0 failures. rust
Prune — Intelligent X/Twitter following manager. Phoenix scorer with 9-signal bot detection, LLM classification, engagement analysis. Web UI + CLI. python
Flywheel — Autonomous agent swarm. Oracle generates intent via semantic gap analysis (Gemini embeddings), triage prioritizes, workers execute, dispatcher auto-scales on RunPod. python + rust
Panopticon — Terminal UI for the Omega Kernel built on FrankenTUI. Real-time subsystem telemetry, agent status, belief state gauges. rust · frankentui
Court Engine — Adversarial multi-model debate simulator. Fan a prompt to 7 frontier models, synthesize agreement and divergence, sign the verdict with Ed25519. rust + typescript
Memori DB — Engram storage with multimodal embeddings. FrankenSQLite backend, 768-dim MRL vectors via Gemini Embedding 2. Semantic search over memory. rust

Key Decisions

Ed25519 over JWT — tokens prove identity, signatures prove what happened
POMDP belief states over rule engines — uncertainty is the feature, not a bug
FrankenSQLite over Postgres — bare metal doesn't need a cluster
FrankenSearch over Elasticsearch — two-tier hybrid search, no JVM
Adversarial debate over single-model consensus — disagreement reveals truth
768-dim MRL over 3072 full — 96% quality at 25% storage
Cosine similarity over L2 distance — direction matters more than magnitude

Beliefs

Every AI decision should produce a cryptographic receipt. If you can't verify it, it didn't happen.
The best infrastructure is the infrastructure you own. Cloud is someone else's bare metal with a markup.
Multi-model adversarial debate beats RLHF for alignment. Disagreement is signal, consensus is noise.
Small, fast binaries. The Omega Kernel is 3.4MB. If your agent framework needs a container, you've already lost.
Speed compounds. Ship daily or don't ship at all.