Medical Device Specialist · iOS Developer · Software Engineer
Gunnar Hostetler
I have 4 iOS apps on the App Store because it's fun. My latest app is an on-device Apple Intelligence RAG engine that runs completely offline. Solving problems is my passion, and I build the solutions I wish existed.
About Me
Healthcare operations by day.
AI developer by obsession.
I work full-time as a medical device specialist at the VA in Palo Alto. I build at night and on weekends because this is what my brain wants to do.
The Build Journey
I wanted an iOS client for OpenAI's Assistants API. OpenAI never made one. So I figured it out and built OpenAssistant.
Then I wanted to do RAG with Pinecone on my phone. So I built OpenCone.
Then Assistants got deprecated and I needed the new Responses API with all the tools. So I built OpenResponses. It passed App Review on first submission.
Then I thought "why not make my own RAG engine?" because I've found myself obsessed with how these systems work, and what AI can help me build (based off of my other 3 apps). Things have gotten much easier over time, especially with the agentic era. I decided to follow my curiosity, and OpenIntelligence was born — 102 services, 29-step pipeline, on-device embeddings, Apple Intelligence. It's the most complex thing I've built and I'm genuinely fascinated by it. It runs IN AIRPLANE MODE!!
Then I wanted to see if AI could reinvent how people discover opportunities. So I built LinkedOut — a full-stack matching platform. It aggregates from 5 sources, scores with LLMs, and lets you swipe through the best matches on iOS.
On the Python side, I built PlaudBlender to turn my daily Plaud voice recordings into a searchable knowledge graph with Gemini AI, Qdrant vector search, and an 11-tool MCP server.
I've figured things out this far, and I trust I'll keep figuring them out as I go.
How I Build
I see a problem, I figure out how it should work, and I don't stop until it's on the App Store. The code comes from models and the vision is mine.
Build Activity
What I Own
The Stack
Healthcare Ops
Project Stratigraphy
Each ring represents a commit. Hover to explore the commit history.
My Apps
OpenResponses
OpenResponses is a native iOS client for OpenAI's Responses API. It runs on iPhone iOS 17+, iPad, and Mac.
- Models: Supports 15+ models including GPT-5.2, GPT-5.1, GPT-4.1, and o3/o1 series. Reasoning models display thinking process; effort is configurable.
- Tools: Code Interpreter (sandboxed Python), Web Search (with citations), File Search (vector stores), and Image Generation (gpt-image-1).
- File Handling: DocumentPicker accepts 43+ formats including PDF, DOCX, XLSX, images, and code. Automatic conversion for unsupported formats.
- MCP Integration: Connects to Notion, Dropbox, Gmail, etc. via Model Context Protocol. Supports custom remote servers.
- Configuration: Granular control over parameters (temperature, top_p, reasoning effort). Prompt Library for saving configs. Request Inspector for raw payloads.
- Privacy: API keys in Keychain. Local JSON conversation storage. No default analytics. No accounts.
- Pricing: Free app. BYOK (Bring Your Own Key) model paying OpenAI directly.
- Audience: Developers, researchers, and power users needing parameter-level control.
OpenIntelligence
OpenIntelligence builds an offline knowledge base on your iPhone. It imports PDFs, images, code, and more, extracting text via high-res OCR and native parsers.
- Ingestion: Content is split into logical sections and indexed locally using both keyword and neural search indexes.
- Retrieval Pipeline: 29-step process with query expansion, hybrid search (text + embeddings), cross-encoder reranking, and contextual compression. Best evidence is prioritized for on-device AI.
- Verification Gates: Seven gates (A-G) per response validate retrieval confidence, evidence coverage, numeric sanity, contradictions, semantic grounding, quote faithfulness, and generation quality. Critical gates trigger full abstention.
- Autonomous Tools: 8 agentic @Tool functions handle searching, summarizing, file comparison, pattern counting, corpus stats, and related content discovery.
- Quality Modes: Standard, Deep Think, and Maximum using Self-RAG 2.0.
- Supported Formats: PDF (native + scanned), Word, Excel, PowerPoint, Markdown, CSV, RTF, plain text, 16 code languages, images, and A/V transcription.
- Privacy: All 102 services run entirely on-device. No accounts, uploads, or tracking. Zero-retention.
OpenCone
OpenCone is a retrieval-augmented generation (RAG) app for iOS that indexes your documents and answers questions using AI.
- Processing pipeline: Documents import via the system file picker. The app copies files to local storage with bookmark persistence. Duplicate detection uses SHA256 hashing. Text extraction runs locally—PDFKit for PDFs, Vision OCR for images. A recursive text splitter creates chunks respecting document structure. Chunks embed via OpenAI (text-embedding-3-large, 3072 dimensions, batches of 50). Vectors store in your Pinecone index.
- Search modes: Semantic search, Hybrid search (requires dotproduct index), and Reranking (BGE, Cohere, Pinecone).
- Answer generation: Queries embed, retrieve top-k results, and pass to OpenAI's Responses API. Answers stream via SSE. Models include GPT-5.2 (400K context), GPT-4o, and o-series. Reasoning models support effort levels.
- AI tools: Code Interpreter runs Python for charts, calculations, and data analysis. Web Search retrieves current information beyond your documents. Both are optional toggles.
- Supported formats: PDF, DOCX, DOC, TXT, RTF, HTML, CSS, Markdown, JSON, XML, CSV, TSV, Python, JavaScript, PNG, JPEG, GIF, TIFF, BMP.
- Data handling: On-device: file access, extraction, chunking, deduplication. Cloud: text chunks to OpenAI, vectors to Pinecone. No third-party analytics.
- Resilience: Retry logic with exponential backoff. Circuit breaker opens after 2 consecutive failures. Rate limiting at 100ms. 30-second watchdog for stalled streams. 100MB file limit.
- Pricing: BYOK - bring your own API keys. No subscription.
- Audience: Researchers, legal professionals, developers, knowledge workers.
OpenAssistant
OpenAssistant is an iOS app that provides access to the OpenAI Assistants API. It allows you to create, configure, and interact with AI assistants from your iPhone or iPad.
- Assistants & Models: Create assistants using GPT-4o, GPT-4.1, or O-series models. Configure parameters like temperature, top_p, and reasoning_effort.
- Tools: Supports Code Interpreter for executing Python and File Search for retrieving information from uploaded documents.
- File Handling: Create vector stores and upload files (PDF, DOCX, CSV, code, images). Configurable chunk size and overlap for retrieval.
- Chat Experience: Thread-based sessions with 2s polling interval. Markdown rendering. History saves locally by thread ID.
- Data Handling: API keys and history stored locally on-device. Direct connection to OpenAI—no third-party servers.
- Reliability: Robust networking with automatic retries (up to 3 attempts) and increasing wait times.
- Pricing: Free and open-source (MIT). BYOK - pay OpenAI directly.
- Audience: Developers, researchers, and students managing OpenAI assistants on mobile.
Other Projects
LinkedOut
AI-powered matching platform — aggregates from 5 sources, scores with LLMs (Gemini + OpenAI fallback), and presents swipeable cards on iOS. Two-tier scoring pipeline with anti-sycophancy prompts, Why Matrix, and configurable weights. FastAPI backend on Docker/Render.
PlaudBlender
Voice recordings → knowledge graph. Ingests Plaud transcripts via OAuth, processes through Gemini AI (cleaning, event extraction, sentiment), indexes to Qdrant vector DB, and visualizes via interactive Dash UI with timeline, knowledge graph, semantic search, and stats. 11-tool MCP server for ChatGPT integration. 91 tests.
Professional Experience
Independent Builder
Building the Open- Series & Beyond
Nov 2023 – PresentLeveraging LLMs to build tools. Since my first commit in late 2023, I've used Foundation Models to generate, refine, and ship complex native iOS applications and Python backends. Driven by a passion for exploration, I orchestrate systems involving RAG, Agents, and on-device intelligence to close the gaps I see in existing software.
- LinkedOut: Full-stack AI matching platform — SwiftUI iOS app + FastAPI backend with LLM-powered scoring (Gemini + OpenAI), 5 data source integrations, and swipe-based UI.
- PlaudBlender: Python knowledge pipeline transforming Plaud voice recordings into a searchable knowledge graph with Gemini AI, Qdrant vector DB, and an 11-tool MCP server.
- OpenResponses: Production successor to my Assistants client with Responses API streaming, Computer Use, MCP connectors, and core tool coverage.
- OpenIntelligence: 102-service on-device RAG engine running on Apple Intelligence with 29-step pipeline, Metal GPU vector search, and 7 verification gates.
- OpenCone: App Store RAG pipeline that pairs Pinecone serverless indexes with OpenAI Responses answers and detailed ingestion analytics.
- OpenAssistant: Legacy SwiftUI client covering Assistants, Threads, Runs, and vector stores before the Responses API launched.
OnSite Specialist
Stryker | VA Palo Alto Health Care System
Aug 2022 – PresentSole technical specialist supporting Stanford surgical teams with complete autonomy. Bridging the gap between complex medical technology and clinical workflows in high-pressure surgical environments.
- Intraoperative Support: Provide real-time technical support for General, Urology, ENT, and Thoracic surgeries, serving as the subject matter expert for advanced medical devices.
- Operations & Procurement: Manage end-to-end procurement and inventory for surgical supplies, forecasting needs based on surgical schedules to prevent case delays.
- Technical Liaison: Act as the primary bridge between surgeons, nursing staff, and biomedical engineering, translating technical constraints into clinical solutions.
- Compliance & Safety: Maintain strict HIPAA compliance and government clearance (PIV) while managing sensitive case data and equipment protocols.
Data Analyst
Above and Beyond Family Recovery Center
Jan 2020 – Jan 2021Managed and validated patient-facing operational data in a clinical environment where accuracy, confidentiality, and process reliability mattered.
- Data Quality: Managed and validated data files for roughly 100 patients per week, ensuring accuracy and compliance.
- Process Improvement: Streamlined data handling workflows to improve operational efficiency and reliability.
- Confidentiality: Maintained strict care while handling sensitive diagnoses, insurance data, and internal records.
Utilities Locator
Utility Resource Group, LLC
Dec 2016 – Dec 2020Field operations role built around precision, safety, route planning, and independent execution in a high-accountability environment.
- Safety & Precision: Located natural gas lines with specialized equipment while maintaining safety compliance.
- Operational Throughput: Navigated mapping systems and company tools to manage daily work, averaging 24 tickets closed per day.
- Training: Mentored and trained 25 new technicians on safety protocols and field best practices.
Physical Therapy Aide
Athletico
Sep 2015 – Oct 2016Supported patient care workflows while handling scheduling, insurance administration, and front-line clinic operations.
- Patient Support: Assisted physical therapists with patient care and rehabilitation exercises.
- Scheduling: Managed appointment changes, cancellations, and day-to-day clinic coordination.
- Insurance Workflow: Verified, scanned, and processed insurance information across Medicare, worker's compensation, and private insurers.
District Manager
Aqua-Guard Management Inc
Mar 2010 – Sep 2015Early leadership role managing distributed operations, staffing coverage, and on-call execution across multiple sites.
- People Management: Supervised seven managers and 31 lifeguards across nine pools.
- Coverage & Reliability: Ensured all shifts were covered, stepping in directly when required.
- Operations: Built daily routes and managed pool maintenance through chemical administration and round-the-clock support.
Education
Western Illinois University
Aug 2010 – Dec 2014Bachelor's Degree in Kinesiology and Exercise Science.
Get In Touch
Let's Connect
I'm always interested in new opportunities and collaborations, especially in iOS development and AI integration. Feel free to reach out if you'd like to discuss a project, explore potential partnerships, or just say hello!