By Innershell
The Magic Lizard is an AI-native enterprise service bus for healthcare — currently in active development. The architecture is proven, the vision is clear, and we're looking for a small number of healthcare teams to build it with us.
The Method
Traditional integration engines demand complex visual canvas mapping. The Magic Lizard inverts that model — you describe what you need, the AI writes and validates the code.
Paste a sample payload and type what transformation you need in plain language. No schemas required to get started.
The LLM orchestrator produces a TypeScript transform plus a full assertion suite — including edge cases like missing fields and extreme values.
An ephemeral Deno subprocess runs all tests in isolation. Failures feed back to the LLM for automatic repair — no human intervention needed.
Once all tests pass, one click pushes the verified artifact to the Go runtime. Zero downtime. Every message logged to the immutable WAL audit trail.
The Self-Healing Loop
The AI execution loop never leaves you with broken code. Every failed test is a structured trace that becomes the next prompt — closing the loop automatically.
The "Deploy" action unlocks. The verified artifact is pushed to the Go runtime with no downtime.
The full stack trace is sent back to the LLM as a structured prompt. The engine iterates — up to N rounds — before surfacing for human review.
Capabilities
Clinical data is messy. HL7 segments go missing. Patient IDs change shape across systems. The Magic Lizard was designed for these conditions from day one.
Describe transformations in plain English. The LLM orchestrator converts intent to production TypeScript with structured output guarantees — no ambiguity, no hand-coding.
Every transform runs in an isolated Deno subprocess with tightly scoped OS capabilities. No transform can reach outside its defined network or file permissions.
SQLite WAL mode guarantees every raw payload is durably written before execution. Point-in-time recovery and full lineage inspection for every message, always.
The AI-generated TypeScript surfaces in a live Monaco editor. Review, tweak, or override any generated code before testing — full developer control, zero ceremony.
Beyond user-provided examples, the engine auto-generates 3+ adversarial edge cases: missing optional keys, special characters, maximum-length fields, and malformed segments.
SSE-streamed ingestion activity flows directly into a scannable table. Inspect exact input/output payloads of any historical transaction without leaving the dashboard.
Under the Hood
Designtime and runtime are fully decoupled. Build and validate in the Next.js studio; execute at high throughput in the Go daemon.
Next.js · Monaco Editor · Tailwind CSS
Structured JSON output · Prompt chaining · Auto-heal loop
Ephemeral Deno subprocess · deno test · Scoped OS caps
Single-binary daemon · MLLP + HTTP + SFTP channels
Isolated Deno workers · stdin/stdout IPC · Per-message execution
Pre-execution write · ACID guarantees · Point-in-time recovery
The Go runtime compiles to a single static binary. Deploy on-premises inside your data center, on a locked-down VM, or across distributed cloud nodes — no runtime dependencies to manage.
Every inbound payload is durably written to the WAL before any transform executes. A process crash mid-flight never results in lost or half-processed data.
Each transform runs in a separate Deno subprocess. A runaway script — infinite loop, memory spike, unhandled exception — cannot affect any other active channel.
For HIPAA-covered environments, the entire stack runs air-gapped. No patient data leaves your network. The LLM call happens at designtime only, against sample data you control.
Protocols & Standards
From legacy MLLP TCP streams in plasma donation centers to modern FHIR R4 REST APIs, The Magic Lizard handles inbound and outbound channels natively.
A note on where we are
The Magic Lizard is not a finished product — it's a fully-designed, actively-developed platform seeking its first design partners. That means you get direct input into what gets built, priority access when it ships, and a pricing structure that reflects the trust you're placing in us. In exchange, we ask for honest feedback and a real integration challenge to build against.
Build It With Us
The architecture is designed, the technology is chosen, and development is underway. We're looking for a small number of healthcare teams ready to co-build this with us — and get a platform shaped around their actual integration pain.
Contact
Reach out to start a conversation. There's no sales pitch here — just a direct discussion about your integration challenges and whether this is something worth building together.
You bring a real healthcare integration problem. We bring the architecture, development capacity, and AI-first approach. We build the parts that matter to you first, you give us honest feedback, and we both end up with something better than either of us would have built alone.
The Magic Lizard is being built by Innershell — medical software specialists with deep experience in plasma, blood & transfusion systems.
Visit Innershell.com