Generate initial workspace docs from facts.md

7 focused documents for the assistant: company overview, career
history, education, AI/ML services, mobile development, project
portfolio, technical expertise. Generated via /build-workspace skill.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

workspace/ai-and-ml-services.md

@@ -0,0 +1,97 @@
+## Table of Contents
+- LLM Harness — Document Exploration Agent -- line 16
+  - Architecture -- line 22
+  - Capabilities Demonstrated -- line 30
+  - Metrics -- line 37
+- LLM Server — Multi-Provider API Infrastructure -- line 43
+  - Architecture -- line 49
+  - Capabilities Demonstrated -- line 56
+- DSPy Experience -- line 64
+  - Capabilities Demonstrated -- line 74
+- Evaluation Frameworks -- line 81
+  - LLM Harness Evaluation Suite -- line 83
+  - MMLU Benchmarking Framework -- line 87
+  - FOT Intervention Recommender — RAG Evaluation -- line 91
+
+## LLM Harness — Document Exploration Agent
+
+[GitHub Repository](https://github.com/chuckfinca/a-simple-llm-harness)
+
+An open-source, minimal agent loop for exploring document workspaces with tool-calling LLMs. The model receives a single tool — `run_python` — which executes code in a sandboxed container with read-only access to workspace files. No framework, no vector database, no retrieval pipeline. The agent reads documents by writing Python code, accumulates evidence, and produces cited answers.
+
+### Architecture
+
+- **Agent Loop:** LLM call → tool execution → event yielding → trace recording, with streaming support for real-time text deltas
+- **Code-as-Tool:** The LLM writes Python to explore documents rather than relying on chunked retrieval (RAG). This trades latency for dramatically better reasoning — the model can filter, compute, cross-reference, and verify across documents
+- **Citation Pipeline:** Model outputs parseable `[filename: "quoted passage"]` citations; server-side regex extraction and three-tier fuzzy matching (exact → ellipsis segment → 5-word window) produce deterministic footnotes with source verification
+- **Streaming:** Server-Sent Events via Gradio API, custom `StreamedCompletion` class for real-time text deltas
+- **Sandboxed Execution:** Docker containers locally, E2B cloud sandboxes in production
+
+### Capabilities Demonstrated
+
+- Multi-provider LLM abstraction via litellm (OpenAI, Anthropic, Google, and others)
+- Per-turn token counts, latency, and cost tracking with traces uploaded to HuggingFace Dataset repos
+- Evaluation suite covering 10 capability categories (single_fact, multi_doc, comparison, trend_synthesis, cited_analysis, and others) across 5 diverse document workspaces
+- Deployed as a live demo on appsimple.io with persona-based prompting, daily rate limiting, and off-topic gating
+
+### Metrics
+
+- ~1,300 lines of Python (harness library) + ~800 lines (assistant app)
+- 111 commits over 25 days of active development
+- Live on appsimple.io as an interactive demo
+
+## LLM Server — Multi-Provider API Infrastructure
+
+[GitHub Repository](https://github.com/chuckfinca/a-simple-llm-kit/tree/c014c51e0429ba257a8012b2668eaf528a051f2b)
+
+A comprehensive FastAPI-based server providing LLM access for mobile applications through RESTful APIs. Designed as the backend bridge between mobile clients and multiple LLM providers.
+
+### Architecture
+
+- **Backend:** FastAPI, Python, DSPy, Pydantic, Uvicorn
+- **Models:** OpenAI GPT-4, Anthropic Claude, Hugging Face models, Google Gemini via unified interface
+- **Infrastructure:** Docker, Modal.com deployment, Cloudflare tunnels, Prometheus monitoring, GitHub Actions CI/CD
+- **Patterns:** Circuit breaker for API failure handling, pipeline processing for multi-step workflows (image → text → structured data), versioning middleware for program and model tracking
+
+### Capabilities Demonstrated
+
+- Standardized interface for all LLM providers, enabling seamless model switching and failover
+- Multi-step pipeline system enabling workflows like image → text → structured data extraction
+- Contact information extraction pipeline for image processing
+- Comprehensive versioning system for program and model tracking
+- Production-ready deployment with monitoring and performance metrics
+
+## DSPy Experience
+
+Charles has applied DSPy in multiple production systems:
+
+- **AI Property Report Generator:** Built structured AI commentary generation using DSPy signatures across 6 distinct report sections. Architected a novel system (`dynamic_signatures.py`) that injects prompt instructions from Excel cells into DSPy signatures at runtime, allowing non-technical users to perform "prompt engineering" by editing spreadsheet cells. Implemented parallel tab generation with prompt caching — running the first tab serially to warm Anthropic's prompt cache, then executing remaining tabs in parallel with async workers.
+
+- **ContactCapture Backend:** Multi-model extraction pipeline built with DSPy supporting OpenAI GPT-4o-mini, Google Gemini, and others for business card data extraction.
+
+- **LLM Server Pipelines:** DSPy-based pipeline processing integrated into the LLM server for structured data extraction workflows.
+
+### Capabilities Demonstrated
+
+- Dynamic DSPy signature generation from external data sources
+- Custom output processors and `BaseProcessorV2` for Pydantic v2 integration
+- MockLM testing for deterministic pipeline validation
+- Production deployment on Modal.com serverless infrastructure
+
+## Evaluation Frameworks
+
+### LLM Harness Evaluation Suite
+
+Code-based assertions against 10 question categories (single_fact, multi_doc, comparison, trend_synthesis, cited_analysis, and others) with session support for multi-turn exploration across 5 diverse document workspaces (SEC filings, Federalist Papers, Sherlock Holmes, world data, Darwin). Enables systematic comparison of models and prompt strategies.
+
+### MMLU Benchmarking Framework
+
+Custom framework for LLM benchmarking achieving 66.6% MMLU accuracy with Llama 3.1 8B, later improved to 71.1 via DSPy-optimized prompts (vs 68.3 with manual engineering). Published findings on evaluation reproducibility challenges, documenting a 6-point performance gap caused by prompt variations alone.
+
+### FOT Intervention Recommender — RAG Evaluation
+
+[GitHub Repository](https://github.com/chuckfinca/fot-intervention-recommender) | [Live Demo](https://huggingface.co/spaces/chuckfinca/fot-recommender-api)
+
+End-to-end RAG system built and deployed in 5 days with 95%+ test coverage. Transforms student narratives into evidence-based intervention recommendations using FAISS vector search, semantic chunking, and persona-based generation (teacher, parent, principal). Knowledge base curated from 6 authoritative educational research sources.
+
+**Technology:** Python, Sentence Transformers, FAISS, Google Gemini API, FastAPI, Gradio

workspace/career-history.md

@@ -0,0 +1,105 @@
+## Table of Contents
+- Centro de Estudios Científicos — Researcher (Jul 2005 – Dec 2005) -- line 10
+- MixMatchMusic, Ltd. — Founder & CEO (Jun 2006 – Sep 2015) -- line 18
+  - Professional Recommendation: Lisa Strauss -- line 41
+- Dig Dates, Inc. — Chief Technology Officer (Apr 2018 – Oct 2019) -- line 47
+- CDIT LLC @ Spear Education — Mobile Developer → Lead Mobile Developer (Dec 2019 – Jun 2023) -- line 65
+  - Professional Recommendation: Rob Lester -- line 83
+- AppSimple LLC — Founder, Principal Software Engineer (May 2013 – Present) -- line 89
+
+## Centro de Estudios Científicos — Researcher (Jul 2005 – Dec 2005)
+
+**Duration:** 6 months
+
+Charles conducted electrophysiological studies on temperature-sensitive TRPV3 protein ion channels using Xenopus oocyte expression systems and two-electrode voltage clamp (TEV) techniques. When endogenous ion channels in the oocytes presented as noise interference, he adapted the research approach to characterize these channels as a secondary focus.
+
+This position provided foundational exposure to scientific reproducibility practices and rigorous experimental methodology that continues to influence his approach to software development and testing.
+
+## MixMatchMusic, Ltd. — Founder & CEO (Jun 2006 – Sep 2015)
+
+**Duration:** 9 years, 4 months
+
+MixMatchMusic, Ltd. was an angel-backed ($400,000) music technology company that evolved from a pioneering collaborative music platform into a leading mobile app platform provider. Charles co-founded the company with Alan Khalfin (CFO) and Derek Prothro (Sr. Technology Advisor).
+
+**MixMatchMusic Community (2006–2009):** A browser-based collaborative music creation platform that challenged the industry belief that high-fidelity music production could not be moved to the cloud. Musicians worldwide could share, mix, and remix stems directly in the browser. The platform won the **DEMOgod Award** at DEMOfall 2008 (top 10 out of 2,200 applicants) and acquired competitor Mix2r in 2009, adding 6,200+ songs and stems.
+
+**MobBase (2009–2015):** A self-service platform enabling artists, brands, and businesses to create custom mobile applications across iOS, Android, and HTML5. Key achievements:
+
+- Launched **1,000+ mobile apps** reaching **nearly 1.5 million users** with **1,400,000+ app downloads**
+- Notable clients included Pepper, Parkway Drive, and Tila Tequila
+- Secured strategic partnerships with TuneCore, IODA, and BandsInTown
+- Featured partner for Mozilla Marketplace for HTML5 apps
+- Attracted acquisition interest from ReverbNation
+- Generated $60,000 in revenue in 2013
+
+**Key Leadership Achievements:**
+
+- Led the company through a successful strategic pivot from collaborative music platform to mobile app platform, anticipating the mobile revolution
+- When Apple banned template-based apps in September 2011 (threatening 400+ live client applications), executed comprehensive crisis response within 48 hours — proactive client communication, technical assessment, and strategic pivot to client-owned developer accounts, preserving 85%+ of client relationships
+- Managed all aspects of business operations including legal, marketing, partnerships, financial operations, and investor relations
+
+### Professional Recommendation: Lisa Strauss
+
+*Digital Marketing Specialist, February 2013 — Former Direct Report*
+
+> "I feel honored to work for such a visionary as is Charles. It's not easy to take a dream and make it a reality, you have to have guts, perseverance, determination; all exactly what Charles has. I've learned a lot from him both in business and in life, it's been a pleasure to work for him. He's built an amazing team and I look forward to many more projects with everyone!"
+
+## Dig Dates, Inc. — Chief Technology Officer (Apr 2018 – Oct 2019)
+
+**Duration:** 1 year, 7 months
+
+Dig was a native iOS dating application for dog lovers. As CTO, Charles led end-to-end technology strategy and development, architecting the platform from the ground up.
+
+**Key Achievements:**
+
+- Scaled the platform to **25K+ monthly active users** and **3K+ daily active users**
+- Achieved **4.7-star App Store rating** with 3-minute average session duration
+- Architected full-stack serverless backend on Firebase (Firestore, Cloud Functions, Realtime Database, Auth, Storage)
+- Built complex matching engine processing user preferences, geographic proximity, and dog compatibility
+- Integrated real-time chat, multi-provider authentication (Email, Phone, Facebook, Google), and content moderation system
+- Spearheaded Android platform launch, managing outsourced development team to achieve 100% feature parity within 2–3 months
+- Built in-app monetization system using StoreKit
+
+[Dig Dating App Demo](https://x.com/DigDates/status/1147264615073624064)
+
+## CDIT LLC @ Spear Education — Mobile Developer → Lead Mobile Developer (Dec 2019 – Jun 2023)
+
+**Duration:** 3 years, 7 months
+
+Spear Education is the industry leader in dental education and practice management solutions. CDIT LLC was the contracting company through which Charles worked on Spear's mobile applications.
+
+**Role Evolution:** Charles joined as Mobile Developer in December 2019. Over 18 months, he established technical credibility through delivery of complex features and architectural contributions across Spear's mobile portfolio. In July 2021, he was selected by the outgoing Lead Mobile Developer to assume technical leadership of the mobile team.
+
+**Key Achievements:**
+
+- Led mobile development team and established technical direction for **5+ production applications** serving thousands of users
+- **Spear Online** (flagship app): Drove architecture and implementation achieving **4.9 App Store rating** from 740+ reviews, **17-minute average session duration**, and **30% reduction in customer churn**. Built complete offline download system, team management features, customizable home screen, and advanced search/filtering
+- **Spear Practice Solutions:** Engineered core architecture for dental practice intelligence platform including patient follow-up system, dynamic scheduling, and advanced filtering — well-received by early-adopter clients
+- **Chairside & Lobby:** Maintained and enhanced patient education apps deployed in thousands of dental offices across iPad and tvOS, including cross-device content sharing via MultipeerConnectivity
+- Created **SpearUIFramework** and **AuthFramework** — shared Swift frameworks standardizing core functionality across the entire mobile portfolio, accelerating new feature development by an estimated 40%
+- Architected CI/CD pipeline reducing deployment from **~3 hours of manual effort to a 15-minute automated workflow**, managing **50+ builds per week**
+- Mentored developers and conducted technical interviews for mobile engineering positions
+
+### Professional Recommendation: Rob Lester
+
+*Software Engineer at Apple, March 2021 — Former Manager at Spear Education*
+
+> "Charles is a very skilled iOS developer. He is always quick to choose the most challenging tasks available and thrives while completing them. Most, if not all, of the most complicated mobile features at Spear were implemented by or had some involvement from Charles. Charles does not shy away from new technologies either. He is always willing to lend a hand outside of his core skill set and does an exceptional job while doing so. It's an honor to have Charles on my team and look forward to working with him in the years to come."
+
+## AppSimple LLC — Founder, Principal Software Engineer (May 2013 – Present)
+
+**Duration:** 12+ years
+
+AppSimple LLC is a US-based sole-member LLC that Charles owns and operates. Originally founded as a mobile app development shop, the company transitioned into an AI and Mobile development consultancy following Charles's M.S. in AI/ML (2024).
+
+**Key Achievements:**
+
+- Built and published consumer iOS apps including **Guided Mind** (295K+ installs, 4.7 App Store rating, 400+ meditations from 9 guides) and **Bodhi Mind** (100K+ sessions, 4.6 App Store rating)
+- Created **a-simple-llm-harness** ([GitHub](https://github.com/chuckfinca/a-simple-llm-harness)), an open-source agent loop for document exploration deployed as a live demo on appsimple.io — ~1,300 lines of Python, 111 commits over 25 days, evaluation suite across 10 capability categories
+- Built comprehensive **LLM server** ([GitHub](https://github.com/chuckfinca/a-simple-llm-kit/tree/c014c51e0429ba257a8012b2668eaf528a051f2b)) with FastAPI, multi-provider support, circuit breaker pattern, and pipeline processing for mobile application backends
+- Developed **AI Property Report Generator** — full-stack system transforming Excel financial data and PDF reports into interactive web reports using DSPy, Next.js, and a Git-backed static site generation pipeline
+- Built **ELA Revit Plugin** for automated electrical element placement — dual-process architecture (C#/Python) with 62 commits over 3 weeks, handling 20+ element types with linked model support
+- Contracted as iOS developer for **Mixify** — 314 commits over 3 months, 90,000+ lines of code added, architected video playback system, skeleton loading infrastructure, and reusable UI components
+- Created reusable Swift packages: **ASimpleAuthKit** ([GitHub](https://github.com/chuckfinca/ASimpleAuthKit)), **ASimpleCameraKit** ([GitHub](https://github.com/chuckfinca/ASimpleCameraKit)), **ASimplePurchaseKit** ([GitHub](https://github.com/chuckfinca/ASimplePurchaseKit))
+- Developing **ContactCapture** — SwiftUI app integrating custom AI backend for business card scanning with FlowCoordinator pattern and privacy-first architecture
+- Built **WWOZ Livewire** — official live music guide for a major public radio station with 18K+ installs and 4.5 App Store rating

workspace/company-overview.md

@@ -0,0 +1,34 @@
+## Table of Contents
+- Founding & Mission -- line 8
+- Company Evolution -- line 12
+- Current Focus: AI + Mobile Development -- line 18
+- Company Structure -- line 28
+- Contact -- line 32
+
+## Founding & Mission
+
+AppSimple LLC was founded by Charles Feinn in May 2013 as an app development shop focused on creating iPhone and Android apps for internal and external clients. Following Charles's M.S. in AI/ML (2024), the company transitioned into an AI and Mobile development consultancy specializing in LLM integration, data synthesis, and Apple ecosystem development.
+
+## Company Evolution
+
+- **2013–2019:** Pure mobile app development focus, creating consumer iOS apps including Guided Mind (295K+ installs) and WWOZ Livewire (18K+ installs)
+- **2019–2024:** Continued mobile development while Charles expanded technical skills through contract work at Spear Education and graduate studies in AI/ML
+- **2024–Present:** Post-masters transition to AI/ML and Mobile hybrid focus, building infrastructure and packages for broader use
+
+## Current Focus: AI + Mobile Development
+
+Core service areas:
+
+- **LLM Integration:** Multi-provider server infrastructure, agent architectures, and pipeline processing
+- **Data Synthesis:** DSPy-based structured generation, RAG systems, and evaluation frameworks
+- **Apple Ecosystem Development:** iOS, tvOS, and iPadOS applications using Swift, SwiftUI, and UIKit
+- **Desktop Plugin Development:** Revit plugin development with C#/.NET 8 and Python backend integration
+- **Reusable Infrastructure:** Open-source Swift packages (ASimpleAuthKit, ASimpleCameraKit, ASimplePurchaseKit) and Python libraries (a-simple-llm-harness)
+
+## Company Structure
+
+AppSimple LLC is a US-based sole-member LLC owned and controlled by Charles Feinn. The company serves as the vehicle for independent software development work, combining mobile app development with AI/ML capabilities.
+
+## Contact
+
+charles@appsimple.io

workspace/education.md

@@ -0,0 +1,101 @@
+## Table of Contents
+- M.S. in Artificial Intelligence & Machine Learning — CSU Global (2024) -- line 13
+  - Coursework -- line 15
+  - Key Projects -- line 26
+- B.S. in Neuroscience — UCLA (2006) -- line 52
+- Professional Certifications -- line 56
+  - Arc Certified Remote Developer — Arc.dev (February 2025) -- line 58
+  - Neural Networks for Machine Learning — Coursera (January 2017) -- line 62
+- Career Philosophy -- line 68
+- Publications & Thought Leadership -- line 72
+  - Machine Learning & AI Engineering Blog -- line 74
+
+## M.S. in Artificial Intelligence & Machine Learning — CSU Global (2024)
+
+### Coursework
+
+- Foundations of Python Programming & Software Development
+- Software Engineering Principles & Design Patterns
+- Advanced Data Structures & Algorithms
+- Operating Systems & Concurrent Programming
+- Machine Learning
+- Computer Vision
+- Applied Artificial Intelligence & NLP
+- Deep Learning & Advanced Machine Learning Models
+
+### Key Projects
+
+**Generative Adversarial Network (GAN) for Image Generation**
+Implemented a GAN from scratch using TensorFlow to generate images of ships, trained on the CIFAR-10 dataset. Developed both generator and discriminator models with custom training loop and progress visualization.
+*Technologies: TensorFlow, Keras, NumPy, Matplotlib*
+
+**Custom Neural Network with Automatic Differentiation**
+Developed a neural network framework from scratch implementing a Value object for autodiff and backpropagation. Built multi-layer network for binary classification integrated with XGBoost for low-confidence predictions.
+*Technologies: Python, NumPy, Scikit-learn, XGBoost*
+
+**Arithmephi — LLM Fine-Tuning for Numerical Tasks**
+Implemented parameter-efficient fine-tuning (QLoRA) on Microsoft's Phi-2 model to enhance arithmetic counting capabilities. Created synthetic dataset generation pipeline and configured 4-bit quantization with BitsAndBytes for efficient training on single T4 GPU.
+*Technologies: PyTorch, Hugging Face Transformers, PEFT (QLoRA), BitsAndBytes*
+
+**Face Detection and Privacy Protection**
+Developed an OpenCV system for face/eye detection with circular blur privacy protection. Implemented multi-angle detection and DBSCAN clustering for improved accuracy.
+*Technologies: OpenCV, Scikit-learn, NumPy*
+
+**Retroactive Search Trees**
+Implemented partially and fully retroactive binary search trees for historical queries and modifications.
+*Technologies: Python, NumPy*
+
+**Large-Scale Parallel Data Processing**
+Designed a multi-process Python application to sum two 1-billion-line files of floating-point numbers using memory-mapping (mmap), chunking, and concurrent processing for data too large for RAM.
+*Technologies: Python, Multiprocessing, mmap, Concurrent Futures*
+
+## B.S. in Neuroscience — UCLA (2006)
+
+Charles's neuroscience degree included a 6-month electrophysiology research position at Centro de Estudios Científicos in Chile, studying temperature-sensitive TRPV3 protein ion channels using two-electrode voltage clamp techniques.
+
+## Professional Certifications
+
+### Arc Certified Remote Developer — Arc.dev (February 2025)
+
+Certification demonstrating proficiency in remote software development practices and technical skills.
+
+### Neural Networks for Machine Learning — Coursera (January 2017)
+
+**Credential ID:** K3PDTS99XM55
+
+Comprehensive course covering neural network fundamentals and machine learning applications.
+
+## Career Philosophy
+
+Charles's career represents a decades-long pursuit of understanding neural networks and their practical applications. Beginning with neuroscience research, he transitioned to technology seeking greater granularity in understanding how networks process information. His entrepreneurial and mobile development experience provided the technical foundation to eventually pursue AI/ML integration, where he now combines deep software engineering expertise with his original passion for neural network architectures.
+
+## Publications & Thought Leadership
+
+### Machine Learning & AI Engineering Blog
+
+**Platform:** [chuckfinca.github.io/machine_learnings](https://chuckfinca.github.io/machine_learnings)
+**Focus:** LLM evaluation, prompt engineering, AI development workflows
+
+#### "Towards a Better Workflow" (January 2025)
+
+[Read article](https://chuckfinca.github.io/machine_learnings/posts/25-01-22-towards-a-better-workflow/)
+
+Detailed exploration of AI-assisted development workflows using Claude for software engineering tasks. Covers AI development workflow, code generation, project management, and git integration.
+
+#### "Beyond Manual Prompts: Using DSPy to Match Meta's MMLU Scores" (December 2024)
+
+[Read article](https://chuckfinca.github.io/machine_learnings/posts/24-12-14-beyond-manual-prompts/)
+
+Deep dive into programmatic prompt optimization using the DSPy framework to achieve state-of-the-art LLM evaluation results. Achieved 71.1 MMLU score vs 68.3 with manual engineering, demonstrating DSPy's effectiveness.
+
+#### "Challenges in Reproducing LLM Evaluation Results" (October 2024)
+
+[Read article](https://chuckfinca.github.io/machine_learnings/posts/24-10-03-challenges-in-reproducing-llm-evaluation-results/)
+
+Case study investigating the complexities of LLM evaluation reproducibility. Highlighted a 6-point performance gap due to prompt variations, emphasizing evaluation challenges in the field.
+
+#### "Creating an Evaluation Framework" (September 2024)
+
+[Read article](https://chuckfinca.github.io/machine_learnings/posts/24-09-02-creating-an-evaluation-framework)
+
+Building a custom framework for LLM benchmarking with focus on MMLU evaluation. Created a reusable evaluation framework, achieving 66.6% MMLU accuracy with Llama 3.1 8B.

workspace/mobile-development.md

@@ -0,0 +1,192 @@
+## Table of Contents
+- Spear Online — Flagship Dental Education Platform -- line 28
+  - Key Metrics -- line 34
+  - Technical Highlights -- line 41
+- Spear Practice Solutions — Dental Practice Intelligence -- line 50
+  - Key Metrics -- line 54
+  - Technical Highlights -- line 59
+- Dig — The Dog Person's Dating App -- line 66
+  - Key Metrics -- line 72
+  - Technical Highlights -- line 78
+- Guided Mind — Wellness & Meditation App -- line 87
+  - Key Metrics -- line 93
+  - Technical Highlights -- line 99
+- Bodhi Mind — Daily Meditation Practice -- line 107
+  - Key Metrics -- line 113
+  - Technical Highlights -- line 119
+- ContactCapture — AI-Powered Business Card Scanner -- line 126
+  - Technical Highlights -- line 130
+- Mixify — AI Image/Video Generation Platform (Contract) -- line 139
+  - Key Metrics -- line 143
+  - Technical Highlights -- line 149
+- Swift Packages -- line 156
+  - ASimpleAuthKit -- line 158
+  - ASimpleCameraKit -- line 166
+  - ASimplePurchaseKit -- line 174
+- CI/CD & DevOps for Mobile -- line 182
+
+## Spear Online — Flagship Dental Education Platform
+
+[App Store](https://apps.apple.com/us/app/spear-online/id1461517174)
+
+Charles drove the architecture and implementation of Spear Education's flagship iOS application as primary developer and eventual team lead. The app provides on-demand continuing education for dental professionals with video streaming, offline downloads, progress tracking, and CE credit tracking.
+
+### Key Metrics
+
+- **4.9 App Store rating** from 740+ reviews
+- **17-minute average session duration**
+- **30% reduction in customer churn** through improved mobile experience
+- Standardized core functionality across **5 company mobile applications**
+
+### Technical Highlights
+
+- **Offline Download Engine:** Complete system using URLSession background tasks with concurrent download management, progress tracking, user-configurable cellular settings, and local notifications
+- **Team Management System:** "Assign to Team" feature with RESTful batch operations and GraphQL backend integration for team progress tracking
+- **Advanced Search:** Multi-tabbed interface with dynamically-fetched categories from GraphQL API and iCloud-synced recent searches
+- **Customizable Home Screen:** Reorderable content carousels with iCloud-synced preferences
+- **Rich Video Player:** AVKit integration with Picture-in-Picture, variable playback speeds, and detailed analytics tracking
+- **Technology:** Swift, UIKit, GraphQL (Apollo), REST (Alamofire), Core Data, AVKit, Firebase, Fastlane, GitHub Actions
+
+## Spear Practice Solutions — Dental Practice Intelligence
+
+Charles engineered the core architecture and features of the SPS iOS application, a practice intelligence tool for dental teams providing daily scheduling, patient follow-up management, and integrated communication tools (Email, SMS, Call).
+
+### Key Metrics
+
+- Well-received by early-adopter clients for design and workflow efficiency
+- Established scalable architecture blueprint for future company mobile projects
+
+### Technical Highlights
+
+- **Patient Follow-up System:** GraphQL mutations and queries for creating, assigning, and tracking follow-up tasks with optimistic UI updates
+- **Advanced Filtering Engine:** Multi-faceted filtering with pre-built filters, provider filtering, and patient status sorting
+- **Automated CI/CD:** Custom scripts for GraphQL schema download and Swift code generation from Apollo toolchain
+- **Technology:** Swift, UIKit, GraphQL (Apollo), REST (Alamofire), Core Data, Firebase, Fastlane
+
+## Dig — The Dog Person's Dating App
+
+[App Demo](https://x.com/DigDates/status/1147264615073624064)
+
+As CTO of Dig Dates, Inc., Charles architected and built the complete iOS dating application and its Firebase backend from the ground up.
+
+### Key Metrics
+
+- **25K+ monthly active users**, 3K+ daily active users
+- **4.7-star App Store rating**
+- **3-minute average session duration**
+
+### Technical Highlights
+
+- **Matching Engine:** Multi-stage algorithm filtering by gender preferences, dog ownership, geographic proximity, and age range, orchestrated through Firebase Cloud Functions
+- **Real-Time Chat:** Integrated ChatSDK with Firebase Realtime Database for thousands of concurrent conversations
+- **Multi-Provider Auth:** Email, Phone, Facebook, and Google Sign-In with account linking and duplicate detection
+- **Content Moderation:** Image management pipeline with administrative review system
+- **Android Launch:** Directed outsourced team to achieve 100% feature parity within 2–3 months
+- **Technology:** Swift, UIKit, Firebase (Firestore, Cloud Functions, Realtime Database, Auth, Storage), Alamofire, StoreKit, Fastlane
+
+## Guided Mind — Wellness & Meditation App
+
+[App Store](https://apps.apple.com/app/guided-mind/id672076838)
+
+Charles architected and implemented the complete application from concept to App Store launch. The app provides a library of over 400 meditations from 9 professional guides with streaming, offline listening, and subscription management.
+
+### Key Metrics
+
+- **295K+ total installs**
+- **100,000+ user sessions**
+- **4.7 App Store rating**
+
+### Technical Highlights
+
+- **Content Delivery:** Firebase Realtime Database for content synchronization, Firebase Storage for audio streaming, custom `FirebaseToCoreData` layer for real-time updates
+- **Offline System:** Track downloading with disk space management and local playback
+- **Monetization:** Multi-tier subscription model (monthly, annual, lifetime) with 14-day free trial via StoreKit
+- **Spanish Localization:** UI elements and interface text localized for Spanish-speaking users
+- **Technology:** Objective-C, UIKit, AVFoundation, Core Data, Firebase, StoreKit
+
+## Bodhi Mind — Daily Meditation Practice
+
+[App Store](https://apps.apple.com/app/bodhi-mind/id1260761658)
+
+A wellness application featuring "Sitting With Bodhi," a daily guided meditation series combining a ~10-minute guided track with a user-configurable silent timer. Includes a library of nearly 300 meditations by Bodhipaksa.
+
+### Key Metrics
+
+- **100,000+ user sessions**
+- **13K+ first-time users**
+- **4.6 App Store rating**
+
+### Technical Highlights
+
+- **Daily Meditation Engine:** Complex state management synchronizing AVPlayer audio playback with custom timer for silent meditation portion, including background chime notifications
+- **Multi-Provider Auth:** Firebase Auth with Email, Google, Facebook, and Apple Sign-In via FirebaseUI
+- **HealthKit Integration:** Logging completed sessions to Apple Health as Mindful Minutes
+- **Technology:** Objective-C, UIKit, AVFoundation, Firebase, StoreKit, HealthKit
+
+## ContactCapture — AI-Powered Business Card Scanner
+
+Charles is architecting a SwiftUI-based iOS application for scanning and extracting contact information from images using a custom AI/ML backend built with FastAPI and DSPy.
+
+### Technical Highlights
+
+- **FlowCoordinator Pattern:** `Action` protocol with concrete actions (`ProcessScanAction`, `ShowReviewAction`, `FinalizeAndEscrowAction`) for testable, decoupled workflows
+- **EscrowManager:** Local persistence system preventing data loss on network failure, automatically re-queueing failed saves
+- **Privacy-First Architecture:** All data-sharing opt-in by default, just-in-time permission requests, granular tiered privacy controls
+- **Accessibility:** Dynamic Type scaling, VoiceOver support with contextual labels and hints
+- **Localization:** Localizable.xcstrings with NSLocalizedString for multi-language readiness
+- **Technology:** Swift, SwiftUI, AVFoundation, Contacts framework, Combine, Firebase, FastAPI/DSPy backend
+
+## Mixify — AI Image/Video Generation Platform (Contract)
+
+Charles was contracted as an iOS developer for Mixify, an AI-powered creative platform for image and video generation using art styles, meme templates, and face-swap technology.
+
+### Key Metrics
+
+- **314 commits** over 3 months
+- **90,000+ lines of code** added, 22,000+ refactored/removed
+- Video playback system handling concurrent initialization across 1–3 CPU cores
+
+### Technical Highlights
+
+- **Video Playback System:** Semaphore-based throttling, memory leak fixes via weak reference patterns, automatic background/foreground lifecycle management, scroll freeze detection (>50ms blocks)
+- **Skeleton Loading:** Diagonal shimmer animations with centralized `SkeletonShimmerCoordinator` and `SkeletonConfigurable` protocol for app-wide consistency
+- **ImageDisplayVC Refactor:** Consolidated 3 separate view controllers into single reusable component using presenter pattern with view adapters
+- **Technology:** Swift, UIKit, AVFoundation, AVKit, Kingfisher, Firebase, RevenueCat
+
+## Swift Packages
+
+### ASimpleAuthKit
+
+[GitHub](https://github.com/chuckfinca/ASimpleAuthKit)
+
+Production-ready Swift Package providing Firebase Authentication integration for SwiftUI applications. Handles Email/Password, Google Sign-In, Apple Sign-In, with biometric authentication (Face ID/Touch ID) support. Manages 7 critical authentication states with automated handling. Reduces authentication implementation time by 2–3 weeks per project.
+
+**Technology:** Swift, SwiftUI, Firebase Auth, Google Sign-In SDK, Keychain Services, LocalAuthentication
+
+### ASimpleCameraKit
+
+[GitHub](https://github.com/chuckfinca/ASimpleCameraKit)
+
+Swift Package providing camera functionality with proper orientation handling, image processing, and SwiftUI integration. Solves the notoriously difficult iOS image EXIF orientation problem.
+
+**Technology:** Swift, SwiftUI, AVFoundation, Core Image
+
+### ASimplePurchaseKit
+
+[GitHub](https://github.com/chuckfinca/ASimplePurchaseKit)
+
+Lightweight StoreKit 2 wrapper with async/await API for fetching products, making purchases, checking entitlements, and managing transaction history. Supports subscriptions, non-consumables, consumables, and promotional offers (iOS 17.4+). Zero external dependencies.
+
+**Technology:** Swift, SwiftUI, StoreKit 2, iOS 16.4+/macOS 13.3+
+
+## CI/CD & DevOps for Mobile
+
+At Spear Education, Charles took full ownership of mobile DevOps infrastructure, architecting a complete CI/CD pipeline from the ground up:
+
+- Reduced deployment from **~3 hours of manual effort to 15-minute automated workflow**
+- Managed **50+ builds per week** across all iOS and tvOS projects
+- Built interactive QA system with comment-triggered workflows generating download links and QR codes on PRs
+- Implemented path-based triggers and dependency caching for monorepo management with XcodeGen
+- Optimized self-hosted macOS runners with custom Ruby compilation and caching
+
+**Technology:** Fastlane, GitHub Actions, XcodeGen, SwiftLint, Xcov, CocoaPods

workspace/project-portfolio.md

@@ -0,0 +1,133 @@
+## Table of Contents
+- LLM Harness — Document Exploration Agent -- line 13
+- LLM Server — Multi-Provider API Infrastructure -- line 26
+- AI Property Report Generator -- line 39
+- Spear Online — Flagship Dental Education Platform -- line 50
+- Dig — The Dog Person's Dating App -- line 63
+- MobBase — Self-Service Mobile App Platform -- line 76
+- Guided Mind — Wellness & Meditation App -- line 87
+- ContactCapture — AI-Powered Business Card Scanner -- line 100
+- FOT Intervention Recommender — AI Educational RAG System -- line 111
+- ELA Revit Plugin — Automated Electrical Element Placement -- line 124
+
+## LLM Harness — Document Exploration Agent
+
+[GitHub](https://github.com/chuckfinca/a-simple-llm-harness) | [Live Demo](https://appsimple.io)
+
+An open-source, minimal agent loop for exploring document workspaces with tool-calling LLMs. Instead of chunking and embedding documents (RAG), the LLM reasons about them directly through code execution in sandboxed containers, producing cited answers with verified source references.
+
+**Technology:** Python, litellm, Docker, E2B cloud sandboxes, Gradio, Server-Sent Events, HuggingFace Spaces
+
+**Metrics:**
+- ~1,300 lines of Python (harness) + ~800 lines (assistant app)
+- 111 commits over 25 days of active development
+- Evaluation suite covering 10 capability categories across 5 document workspaces
+
+## LLM Server — Multi-Provider API Infrastructure
+
+[GitHub](https://github.com/chuckfinca/a-simple-llm-kit/tree/c014c51e0429ba257a8012b2668eaf528a051f2b)
+
+Comprehensive FastAPI-based server providing unified LLM access for mobile applications. Features pipeline processing for multi-step workflows, circuit breaker patterns for resilient API failure handling, and multi-provider support with seamless model switching.
+
+**Technology:** FastAPI, Python, DSPy, Pydantic, Docker, Modal.com, Cloudflare tunnels, Prometheus, GitHub Actions
+
+**Metrics:**
+- Production-ready deployment with Modal.com and Cloudflare integration
+- Supports OpenAI GPT-4, Anthropic Claude, Google Gemini, and Hugging Face models
+- Pipeline processing architecture supporting image → text → structured data workflows
+
+## AI Property Report Generator
+
+An enterprise-grade, full-stack automated reporting system transforming raw Excel financial data and PDF ownership reports into professional, interactive web-based property reports. Uses a Git-backed Static Site Generation pipeline — generating immutable TypeScript code (`data.ts`) committed to version control, triggering atomic deployments with perfect version history and zero runtime database latency.
+
+**Technology:** Next.js 16, React 19, TypeScript, Tailwind CSS 4, Recharts, FastAPI, DSPy, Claude Sonnet, Modal.com, SheetJS, HyperFormula, Supabase Postgres, Vercel, GitHub Actions
+
+**Metrics:**
+- Reduces report generation from manual days to automated minutes
+- Handles ~1,000 distinct financial data field mappings
+- Generates structured AI commentary across 6 report sections using DSPy signatures
+
+## Spear Online — Flagship Dental Education Platform
+
+[App Store](https://apps.apple.com/us/app/spear-online/id1461517174)
+
+Premier on-demand continuing education platform for dental professionals. Provides video streaming, offline downloads, progress tracking, CE credit tracking, team management, and advanced search. Charles served as primary developer and eventual team lead.
+
+**Technology:** Swift, UIKit, GraphQL (Apollo), REST (Alamofire), Core Data, AVKit, Firebase, Fastlane, GitHub Actions
+
+**Metrics:**
+- 4.9 App Store rating from 740+ reviews
+- 17-minute average session duration
+- 30% reduction in customer churn
+
+## Dig — The Dog Person's Dating App
+
+[App Demo](https://x.com/DigDates/status/1147264615073624064)
+
+Native iOS dating application for dog lovers with full-stack Firebase backend. Features matching engine, real-time chat, multi-provider authentication, content moderation, and in-app monetization. Charles served as CTO, architecting the entire platform.
+
+**Technology:** Swift, UIKit, Firebase (Firestore, Cloud Functions, Realtime Database, Auth, Storage), Alamofire, StoreKit, Fastlane
+
+**Metrics:**
+- 25K+ monthly active users, 3K+ daily active users
+- 4.7-star App Store rating
+- 3-minute average session duration
+
+## MobBase — Self-Service Mobile App Platform
+
+A self-service platform enabling artists, brands, and businesses to create custom mobile applications across iOS, Android, and HTML5. Featured real-time content management, automated app store submission, and multi-platform deployment from a single dashboard. Charles founded and led the company (MixMatchMusic, Ltd.).
+
+**Technology:** iOS, Android, HTML5, Adobe AIR, custom CMS, automated deployment tools
+
+**Metrics:**
+- 1,000+ mobile apps launched
+- Nearly 1.5 million users reached
+- 1,400,000+ app downloads
+
+## Guided Mind — Wellness & Meditation App
+
+[App Store](https://apps.apple.com/app/guided-mind/id672076838)
+
+Wellness application featuring 400+ meditations from 9 professional guides with streaming, offline listening, Spanish localization, and subscription management. Charles built the entire application from concept to App Store launch.
+
+**Technology:** Objective-C, UIKit, AVFoundation, Core Data, Firebase, StoreKit
+
+**Metrics:**
+- 295K+ total installs
+- 100,000+ user sessions
+- 4.7 App Store rating
+
+## ContactCapture — AI-Powered Business Card Scanner
+
+SwiftUI-based iOS application for scanning and extracting contact information using a custom FastAPI/DSPy backend. Features a FlowCoordinator pattern for complex multi-step workflows, EscrowManager for offline resilience, privacy-first architecture with opt-in data sharing, and comprehensive accessibility support.
+
+**Technology:** Swift, SwiftUI, AVFoundation, Contacts framework, Combine, Firebase, FastAPI, DSPy
+
+**Metrics:**
+- Sophisticated FlowCoordinator pattern for testable, decoupled workflows
+- Three reusable Swift packages created as byproducts (ASimpleAuthKit, ASimpleCameraKit, ASimplePurchaseKit)
+- Multi-model AI extraction pipeline (GPT-4o-mini, Google Gemini)
+
+## FOT Intervention Recommender — AI Educational RAG System
+
+[GitHub](https://github.com/chuckfinca/fot-intervention-recommender) | [Live Demo](https://huggingface.co/spaces/chuckfinca/fot-recommender-api)
+
+End-to-end RAG system transforming student narratives into evidence-based intervention recommendations for educators. Features semantic chunking, FAISS vector search, and persona-based generation for three stakeholder types (teacher, parent, principal).
+
+**Technology:** Python, Sentence Transformers, FAISS, Google Gemini API, FastAPI, Gradio, Hugging Face Spaces
+
+**Metrics:**
+- Complete system built and deployed in 5 days
+- 95%+ test coverage with pytest suite
+- Knowledge base curated from 6 authoritative educational research sources
+
+## ELA Revit Plugin — Automated Electrical Element Placement
+
+A Revit plugin that automatically places electrical elements in residential apartment projects. Uses a dual-process architecture: C# plugin handles Revit API interaction and WPF UI, while a Python FastAPI backend handles placement planning and business logic.
+
+**Technology:** C#, .NET 8, Autodesk Revit SDK 2025/2026, WPF, CommunityToolkit.MVVM, FastAPI, Python, Pydantic, Serilog
+
+**Metrics:**
+- 62 commits over 3 weeks of active development
+- ~3,200 lines of C# (plugin) and ~570 lines of Python (backend)
+- Handles 20 element types, architected to scale to 60+

workspace/technical-expertise.md

@@ -0,0 +1,154 @@
+## Table of Contents
+- Programming Languages -- line 18
+- Mobile Development -- line 29
+- AI & Machine Learning -- line 46
+- Backend Development -- line 66
+- Frontend / Web Development -- line 82
+- Desktop / Plugin Development -- line 94
+- DevOps & Infrastructure -- line 103
+- Development Tools -- line 120
+- Architecture Patterns -- line 130
+  - Protocol-Oriented Programming (Expert) -- line 132
+  - Domain-Driven Design (Advanced) -- line 136
+  - Event-Driven Architecture (Advanced) -- line 140
+  - Code-as-Tool Agent Pattern (Advanced) -- line 144
+  - Dual-Process IPC (Advanced) -- line 148
+  - MVVM / MVC (Expert) -- line 152
+
+## Programming Languages
+
+| Language | Proficiency | Experience |
+|---|---|---|
+| Swift | Expert | 10+ years |
+| Objective-C | Expert | 8+ years |
+| Python | Proficient | 2+ years (AI/ML focus) |
+| TypeScript | Advanced | Full-stack web applications, type-safe API design |
+| JavaScript | Proficient | Web development, Node.js backends |
+| C# | Intermediate | .NET 8, Revit plugin development with WPF |
+
+## Mobile Development
+
+| Technology | Proficiency |
+|---|---|
+| iOS Development | Expert (12 years) |
+| SwiftUI | Advanced |
+| UIKit | Expert |
+| Core Data | Advanced |
+| AVFoundation | Advanced |
+| Firebase | Expert |
+| StoreKit / StoreKit 2 | Advanced |
+| MultipeerConnectivity | Advanced |
+| HealthKit | Advanced |
+| Core Animation | Intermediate |
+| Contacts Framework | Advanced |
+| Combine | Advanced |
+
+## AI & Machine Learning
+
+| Technology | Proficiency |
+|---|---|
+| LLM Integration (multi-provider) | Advanced |
+| DSPy | Advanced |
+| OpenAI API | Advanced |
+| Anthropic Claude API | Advanced |
+| Google Gemini API | Advanced |
+| Hugging Face (deployment, evaluation) | Intermediate |
+| RAG Systems (end-to-end) | Advanced |
+| Vector Databases (FAISS) | Advanced |
+| Semantic Search | Advanced |
+| Prompt Engineering | Advanced |
+| LLM Evaluation Frameworks | Advanced |
+| Agent Loop Design | Advanced |
+| Pydantic | Advanced |
+| litellm | Advanced |
+| Sentence Transformers | Intermediate |
+
+## Backend Development
+
+| Technology | Proficiency |
+|---|---|
+| FastAPI | Advanced |
+| REST APIs | Expert |
+| GraphQL (Apollo) | Advanced |
+| Firebase (full-stack serverless) | Expert |
+| Node.js | Intermediate |
+| Microservices Architecture | Advanced |
+| Supabase (Postgres, RLS) | Intermediate |
+| Google Cloud Platform | Advanced |
+| AWS S3 | Advanced |
+| Gradio | Advanced |
+| Server-Sent Events (SSE) | Advanced |
+
+## Frontend / Web Development
+
+| Technology | Proficiency |
+|---|---|
+| Next.js (App Router, SSG/SSR) | Advanced |
+| React | Advanced |
+| Tailwind CSS | Advanced |
+| Recharts | Intermediate |
+| SheetJS | Advanced |
+| HyperFormula | Advanced |
+| Vercel (Edge, KV/Blob) | Advanced |
+
+## Desktop / Plugin Development
+
+| Technology | Proficiency |
+|---|---|
+| Autodesk Revit SDK (2025/2026) | Intermediate |
+| WPF (MVVM, CommunityToolkit.MVVM) | Intermediate |
+| .NET 8 | Intermediate |
+| Serilog | Intermediate |
+
+## DevOps & Infrastructure
+
+| Technology | Proficiency |
+|---|---|
+| Docker | Advanced |
+| CI/CD Pipelines | Expert |
+| GitHub Actions | Advanced |
+| Fastlane | Expert |
+| Cloud Deployment (Modal.com, AWS, GCP) | Advanced |
+| HuggingFace Spaces | Advanced |
+| E2B Cloud Sandboxes | Advanced |
+| uv (Python package management) | Advanced |
+| Prometheus | Intermediate |
+| Cloudflare Tunnels | Intermediate |
+| XcodeGen | Advanced |
+| CocoaPods | Advanced |
+
+## Development Tools
+
+| Tool | Proficiency |
+|---|---|
+| Xcode | Expert |
+| Git | Expert |
+| GitHub | Expert |
+| VS Code | Advanced |
+| Terminal / CLI | Advanced |
+
+## Architecture Patterns
+
+### Protocol-Oriented Programming (Expert)
+
+Extensive use across all Swift projects. Charles designed protocol-based architectures for authentication services, network layers, camera systems, purchase management, and UI components. Applied at Spear Education to create testable, modular frameworks shared across 5+ applications.
+
+### Domain-Driven Design (Advanced)
+
+Applied in the AI Property Report Generator with bounded contexts, domain modeling, and event-driven webhook architecture. Used to separate concerns between data pipeline, AI engine, and frontend presentation.
+
+### Event-Driven Architecture (Advanced)
+
+Implemented webhook pipelines and async workflows in the AI Property Report Generator. Designed event-driven notification systems using Firebase Cloud Functions at Dig. Built streaming event architectures for the LLM harness.
+
+### Code-as-Tool Agent Pattern (Advanced)
+
+Designed and implemented in the LLM harness as an alternative to RAG. The LLM receives a single `run_python` tool and reasons about documents by writing and executing code, enabling filtering, computation, cross-referencing, and verification that chunked retrieval cannot match.
+
+### Dual-Process IPC (Advanced)
+
+HTTP bridge architecture between C# Revit plugin and Python FastAPI backend in the ELA Revit plugin, with health checks, port conflict detection, and graceful lifecycle management.
+
+### MVVM / MVC (Expert)
+
+Applied across all mobile projects. Custom MVC with manager-based patterns at Spear Education and Dig. MVVM with CommunityToolkit.MVVM for WPF in the Revit plugin. FlowCoordinator pattern extending MVVM in ContactCapture.