GSI AI / Muno Labs Proposal App

Context

GSI AI is a reference architecture generator for Global System Integrator teams working on healthcare AI deployments. It helps partner teams create proposal-ready architecture collateral without waiting for a senior architect to recreate the same regulated patterns from scratch.

The public app is access-gated, so the OperatorLab case study is designed to stand on its own.

Problem

Healthcare AI architecture is repetitive and high-stakes at the same time. Every engagement needs HIPAA guidance, PHI handling, BAA awareness, cloud-specific infrastructure notes, provider-specific implementation details, risk framing, and a credible rollout plan.

That work is too important to fully improvise and too common to rebuild from zero.

What Shipped

• Inputs for healthcare use case, deployment cloud, AI provider, integration pattern, data classification, and scale tier.
• Provider-agnostic generation that supports Claude and OpenAI ChatGPT.
• Proposal-ready outputs: Mermaid architecture diagram, executive summary, component inventory, data flows, HIPAA checklist, deployment guide, roadmap, risk register, and sample code.
• Streaming sections generated in parallel for better perceived latency.
• In-memory caching for repeated demo configurations.

Architecture

The app uses a React/Vite frontend and a FastAPI backend with Pydantic validation. Generation requests route to the selected provider while preserving provider-neutral product language where the provider choice does not matter.

The architecture deliberately separates deployment cloud from LLM provider, because real enterprise buyers often make those decisions independently.

Key Decisions

• Separate cloud choice from model choice. Healthcare customers may standardize on AWS or Google Cloud separately from Claude or OpenAI, so the workflow keeps those decisions independent.
• Generate sections in parallel. Architecture, compliance, deployment, roadmap, and risk content can stream as separate sections instead of waiting for one monolithic response.
• Make compliance explicit. PHI touchpoints, BAA considerations, logging guidance, and human-review notes are first-class outputs rather than generic disclaimers.
• Use code-based diagrams. Mermaid output makes generated architecture visible, editable, and portable into partner collateral.

AI / Workflow Layer

The workflow codifies the shape of a partner architecture engagement: gather customer context, select deployment constraints, generate a structured reference architecture, and produce collateral a pre-sales or solution architecture team can refine.

The AI output is not treated as final authority. It is treated as the first 80 percent of a regulated architecture package.

Sample Artifact

A strong generated package includes:

• Executive summary for the buyer.
• Architecture diagram with PHI touchpoints.
• Component inventory and data-flow notes.
• HIPAA checklist and BAA considerations.
• Deployment plan with IAM, networking, monitoring, and audit controls.
• Risk register with mitigation owners.
• Provider-specific Python and TypeScript integration examples.

Constraints

• The product must not imply legal, privacy, or security approval.
• Generated content must mark PHI touchpoints and avoid raw PHI in logs.
• Provider-specific guidance must update when the selected provider changes.
• Output needs to be useful in a partner conversation, not just technically plausible.

Tradeoffs

• A healthcare focus narrows the market, but it makes the demo more credible because the compliance burden is concrete.
• Provider-agnostic positioning avoids vendor lock-in language, but it requires more careful prompt and UI design.
• Generated architecture accelerates the first pass, while human review remains mandatory for regulated production decisions.

What I Would Improve Next

• Add downloadable architecture packets for common healthcare use cases.
• Add a comparison view that shows how the same use case changes across AWS/GCP and Claude/OpenAI.
• Add review checkpoints for security, privacy, and clinical stakeholders.

What It Proves

This system shows how to turn senior architecture judgment into a repeatable partner enablement workflow for regulated enterprise AI.