My work involves making complex systems understandable by making their structure, assumptions, and tradeoffs explicit: whether that's how a language model predicts the next token, how an agent discovers what tools it has, or how a community weighs competing options in high-stakes decisions.
I like taking systems that are opaque, implicit, or fragmented and making them inspectable, structured, and decision-capable.
- structural-explainability - designing information structures that remain reliable over time, even when interpretations differ. Explores ideas applicable to science, policy, law, civic infrastructure, digital twins, and AI explainability.
Structured tools for making competing values and policy assumptions visible in high-stakes civic decisions. Multidimensional evaluation projects include:
Explorers are based on the Multidimensional Evaluation Engine: A domain-neutral engine for multidimensional evaluation under explicit policy assumptions.
Hands-on examples of how LLM chatbots (e.g. ChatGPT, Claude, and more) have evolved to get better at repeatedly predicting the next word, using context windows and training text.
- App (Text Prediction) - interactive next-token prediction in GPT-style language models
- App (Text Prediction) Source Repo
- Introduction and Pre-Trained Models - progressive examples showing how pre-trained models with different context windows (e.g. 0, 1, 2, or 3 prior words) and different types of training text (e.g., neutral vs structured) shape next token predictions
Tools and conventions for helping AI agents understand, navigate, and work reliably with external systems. (Uses SKILL.md conventions.)
- Adaptive Conformance Specification (ACS) - foundational behavioral protocol: observe, infer, and conform before generating any artifact
- Adaptive Tool Discovery (ATD) - maps the capabilities of an external tool set
- Adaptive Onboarding (AO) - builds the context to work as a member of a specific team or project (conventions, ownership, norms)
Standards-based tooling and guides for professional Python analytics workflows, using uv, ruff, Zensical, and modern project conventions.
- ⭐applied-computing-foundations - repos, file systems, terminals, and navigation basics
- ⭐GUIDE: pro-analytics-02 - guide to professional analytics using
uvand Python - ⭐datafun-toolkit - PyPi package - privacy-aware logging and diagnostics
A seven-module curriculum for building real-time anomaly detection, signal monitoring, and drift-detection systems in Python:
- cintel-01-getting-started
- cintel-02-static-anomalies
- cintel-03-signal-design
- cintel-04-rolling-monitoring
- cintel-05-drift-detection
- cintel-06-continuous-intelligence
A seven-module curriculum covering text preprocessing, exploration, API data collection, web scraping, and NLP pipelines.
- nlp-01-getting-started
- nlp-02-text-preprocessing
- nlp-03-text-exploration
- nlp-04-api-text-data
- nlp-05-web-documents
- nlp-06-nlp-pipeline
- toy-gpt exploration
- Quad9 DNS Resolver - free service
- Firefox / Settings / Privacy & Security: DNS over HTTPS, enable Max and set to https://dns.quad9.net/dns-query OR add Mullvad VPN and set to Default.
- Chrome / Settings / Privacy & Security / Security / Use secure DNS: toggle on, select Custom, enter https://dns.quad9.net/dns-query
- Android / Settings / Network & Internet / Private DNS: Set Private DNS provider hostname to <dns.quad9.net> OR add Mullvad VPN and set to Automatic.
- Starlink App / Settings / Router tab / Custom DNS: Toggle on, enter Primary: 9.9.9.9 and Secondary: 149.112.112.112 (allow 10 min to apply)
- Windows / Settings / Network & internet / WiFi / Choose the network name / DNS server assignment: click Edit / Toggle on IPV4 / Preferred DNS: 9.9.9.9 DNS over HTTPS: On (automatic template) / Alternate DNS: 149.112.112.112 DNS over HTTPS: On (automatic template) / click Save.
- Verify: https://on.quad9.net
- Proton Mail Plus - email provider, $48/yr
- Proton Pass - password storage (if using Proton Mail)
- BitWarden - password storage
- Firefox Browser with
- Containers
- uBlock Origin
- Settings / Privacy & Security: Standard or Strict Tracking Protection
- Settings / Privacy & Security: UNCHECK Ask to save passwords
- Settings / Privacy & Security: DNS over HTTPS, enable Max and set to https://dns.quad9.net/dns-query
- FUTO Keyboard - mobile keyboard with swipe and voice that sends nothing (still in alpha)
- Mullvad VPN - VPN for 5 devices, $78/yr, Windows 11, Mac, Android
- Verify at https://mullvad.net/en/check
- signin-sheet-builder - when you have too many columns to fit on a landscape page and you want 2 rows per member
- templates - favorite CI/CD GitHub actions, config files, MkDocs, etc.
- chrome_new_tab_ext
- IntroToPython - professional project setup for the textbook examples
- datafun-01-foundations - professional Python foundations
- datafun-02-automation - automation with repetition, ranges, list comprehensions, delays
- datafun-03-analytics - read, process, write text and structured data
- datafun-04-notebooks - explore/present insights using Python and Markdown in a Jupyter notebook
- datafun-05-sql - Python and SQL
- datafun-06-spec - custom EDA project
- datafun-07-regression - linear regression and prediction
- applied-ml-template - copy this to your GH and replace template with yourname
- confusion-matrix-explorer - PyShiny interactive app repo - try the app here.
- ml-04 - classification analysis
- ml-07 - regression analysis
Uses pip and venv with requirements.txt:
- pro-analytics-01
- pro-analytics-01-audio-guides - repo used to generate podcasts with AI (crazy!)
- Listen <-- try the podcasts here
- RockSwap - match-3 game
- (repo) - source code
- RockSwap-Vivid - match-3 game (red-green colorblind safe)
- (repo) - source code
- RockFit - falling blocks game
- (repo) - source code
- Life Expectancy at Birth (1960–2022) - interactive chart to compare life expectancy at birth by country, region, or economic classification.
- Life Expectancy By Location (Bar Chart Race) - YouTube video of the top 30 from 1960-2022
- Minnesota GIS: Viewer
- Civic Interconnect (organization
- ☆Zig Bill Utils - high-performance toolkit for analyzing U.S. legislative bills.
- ☆Py Bill Utils - Python project for analyzing U.S. legislative bills. - uses Zig Bill Utils
- Google NotebookLM - AI-generated summaries, Q&A, and podcasts from your own documents and sources.
- Meta Quest Virtual Reality - immersive games, learning, productivity, and creativity in mixed reality.
- Google Apps Script - automate workflows in Google Drive, Docs, Sheets, Gmail, and more using JavaScript-like scripting.
- Microsoft Quantum Simulator - Try quantum computing in your browser—no special hardware required.
- Learn with Quantum Katas - interactive notebook tutorials for quantum logic and computation.
- Viola the Bird - interactive experiment blending machine learning with cello-inspired play.
- BirdNET – Identify Bird Sounds with ML - Upload audio or use field device to identify birds by sound with bioacoustics AI.
- GitHub Copilot Workspace - AI-powered IDE that understands your repo, suggests tasks, writes code, and walks through implementation plans.
- GitHub Copilot Voice - use your voice to code (without spelling things out)
- More GitHub Next Projects - cutting-edge developer tools being prototyped by GitHub.
- pro-analytics-ai — illustrates how LLMs like ChatGPT work by creating an AI assistant trained on pro-analytics-01. It uses APIs (OpenAI/OpenRouter) to process user prompts and generate natural language responses, but all retrieval and context selection is performed locally based on the source repository text.
- The AI assistant runs locally on a 2 TB SSD machine with 12/24 cores, using API calls (via a free OpenRouter API key).
- The frontend is lightweight, but the backend uses a local retrieval pipeline (ChromaDB + sentence-transformers) that depends on large Python ML libraries, making it too heavy for serverless platforms like AWS Lambda and significant even for AWS EC2.
- Even when calling external APIs (and not hosting full LLM models), generating embeddings and performing document retrieval requires a significant machine learning environment.
- When we build a brain by indexing our own documents, the storage, memory, and processing requirements expand dramatically.
Early work:
- 2020 May Research Paper introducing GPT-3: "Language Models are Few-Shot Learners"
- 2022 Nov OpenAI ChatGPT "research preview. Current link
- 2023 Nov: OpenAI DevDay - first developer conference (GPT-4 Turbo, and more).
- Conferences:
- test page
- mandart-engine-rust
- Comparing input to output
- Example input Frame25 and output Frame25
- smart-sales-analysis-goals
- smart-sales-raw-data
- smart-sales-example
- smart-sales-starter-files
- smart-sales-docs
- smart-sales-olap
- smart-sales-spark
- smart-sales-duckdb-sql
- smart-sales-duckdb-sql-python
- buzzline-01-case - Python generators for streaming pipelines
- buzzline-02-case - Kafka pipelines for streaming data
- buzzline-03-case - Kafka pipelines for streaming structured and semi-structured data
- buzzline-04-case - Kafka pipelines with live visualizations
- buzzline-05-case - Kafka pipelines with data storage
- buzzline-06-world - Simulated historical analysis of streaming data worlds
- buzzline-06-private - Instructor only civic world simulation
- kafka-producer-earthquake - Kafka producer for streaming earthquake data from USGS API
- dc-mailer
- dc-texter
- pro-analytics-apache-starter - keep JDK, JAVA_HOME, etc isolated
World Diabetes Day (WDD) is the world’s largest diabetes awareness campaign reaching a global audience of over 1 billion people in over 160 countries. It is marked every year on 14 November, the birthday of Sir Frederick Banting, who co-discovered insulin along with Charles Best in 1922.
WDD was created in 1991 by International Diabetes Federation (IDF) and the World Health Organization and became an official United Nations Day in 2006 with the passage of United Nations Resolution 61/225.
- MandArt (App Store) - runs on MacOS (v12+) - Released!
- MandArt-Docs
- MandArt-Discoveries
- Color Cube
- MandArt-Web
A SwiftUI app for creating custom art with the Mandelbrot set.
- MandArt: source repo
- MandArt: my fork
- MandArt deploy
- MandArt-Docs: source repo
- MandMath (w/Julia) - Julia enables 150 - 1500+ decimal places
- Color Cube: source repo
- GrandArt: source repo - explore exponenents beyond 2
- MandArt3: source repo - explore MandArt with exponenent 3
- pyshiny-tips-dashboard-express
- pyshiny-penguins-dashboard-express
- pyshiny-penguins-dashboard-core
- cintel-01-getting-started
- cintel-02-app
- cintel-03-data
- cintel-04-reactive
- cintel-05-live-updates
- cintel-stock-response
- fetch-data-weather
- cintel-01-spec
- cintel-02-spec
- cintel-03-spec
- cintel-04-spec
- cintel-05-spec
- cintel-06-spec
- cintel-07-spec
- digital-forensics-resources
- 386-site
- dfir-hash-1
- dfir-hash-2
- dfir-pwsh-forensic-copy
- dfir-pwsh-file-recovery
- df-mod6-mobile
Beginner-friendly - these skills are great for students to explore GitHub, open-source, and Alexa projects. Schedules can be updated with no coding experience - follow a link below and look in lambda/games.json to see how easy it is to update the schedule when a new season begins.
Skills to ask Alexa about upcoming local games in Ely Minnesota. ❄
- TimberwolfTracker-BoysBasketball
- TimberwolfTracker-GirlsBasketball
- TimberwolfTracker-GirlsVolleyball
