The Platform
High-agency AI orchestration and local-first software. Shipped to the App Store, built for real-world clinical and automation workflows.
I am Gunnar Hostetler. I support surgical technology for Stanford surgeons at the VA in Palo Alto, CA. In my free time, I build and design SwiftUI apps, and map systems while partnering with AI - my software engineer - to write and refine the code.
Click on any node in the architectural diagram above to inspect how that processing step operates inside the native Swift engine.
Understanding each phase of on-device RAG is critical to optimizing memory, latency, and context limitations.
Select a node to see technical implementation details, Swift API methods, and algorithmic settings.
No active log trace.
const platform = orchestrate({
model: "local-first when possible",
pipeline: ["ingest", "chunk", "reason", "act"],
interface: "native utility",
failureMode: "visible, recoverable, logged",
});
platform.ship({
bias: "working software",
overhead: "low",
telemetry: "every important click",
});Autonomy in Action
I work inside surgical environments at the VA Palo Alto Health Care System, providing intraoperative support across eight operating rooms. In the OR, technical failure is not an abstract bug—it has immediate stakes. That workflow obsession directly shapes my software.
I am not a software engineer, and I don't write raw syntax by hand—I utilize AI as my software engineer. Instead, I act as the systems architect, mapping out the logic, the schemas, the inputs, and the edge cases. I design local AI systems, PDF parsing pipelines, and direct database storage utilities, using AI to turn those blueprints into working code.
Active product nodes
Each application represents a direct response to a real-world workflow limit. Built in SwiftUI, optimized for the Apple ecosystem, and engineered with granular local-first principles.
01
Clinical Workspace with Local Intelligence
Provider-facing charting workspace for iPhone, iPad, Mac Catalyst, and visionOS. Uses SwiftData, Apple Foundation Models, SMART on FHIR imports, and OpenIntelligence-derived retrieval components for grounded local clinical workflows.
02
Apple Intelligence Agentic RAG
Apple Intelligence-fueled document system with Apple's Foundation Models framework, a local 3B-parameter on-device model path, agentic RAG orchestration, PDFKit extraction, Vision OCR, and evidence-backed answers.
03
Direct Responses API Playground
Native SwiftUI client for /v1/responses with streaming events, reasoning traces, web search, code interpreter, browser automation, remote MCP, and local Keychain-backed API access.
04
Cloud-Hybrid OCR and Vector Search
SwiftUI client that extracts and OCRs files locally with PDFKit and Vision, then sends embeddings and search traffic to OpenAI and Pinecone serverless indexes. Useful for private corpora, but not a fully offline app.
05
Legacy Assistants API Client
Older native dashboard for OpenAI Assistants API threads, runs, files, and vector stores. Kept online for support and reference, but superseded by OpenResponses.
Clean separation
Personal builder portfolio, high-agency creator identity, and professional technical anchor.
Personal Portfolio ->Publisher site for product landing pages, support routes, and review-facing legal infrastructure.
Support Hub ->Interactive hub for AI orchestration, active script experiments, live runtime simulations, and feature testing.
Explore Systems ->Telemetry posture