I'm so glad you found me! I'm Josh and I still love data — only now I get paid to help other people love it too.

These days I'm an AI/ML educator and builder. I teach Generative AI at Lower Columbia College, run an AI consultancy (CaellwynAI) helping about a dozen clients put LLMs and classical ML to work, and have spent the last four-plus years teaching data science and machine learning to adult learners at Coding Dojo, LearningFuze, and Flatiron School. Before all that, I was a preschool teacher for 7 years — known as "The Toddler Whisperer" — and helped start three project-based schools. Turns out the core idea transfers: people (toddlers, bootcamp students, senior engineers) learn best when they feel seen, heard, and respected, and when what they're learning actually matters.

🗣️ Polyglot Forms — multilingual voice-AI form system Real-time voice form-filling using the Gemini Live API with <50ms barge-in latency and support for 18+ spoken languages. HIPAA-aware architecture keeps PII client-side; piloted with a nonprofit serving recent immigrants. FastAPI + WebSockets + React/TypeScript on Google Cloud.

🧠 Long-Memory Character Chatbot — RAG-based conversational AI A novel architecture that keeps effectively unbounded conversational memory while holding active context under 4,000 tokens. Multi-agent pipeline (main conversation, memory summarization, content-safety filtering) with adjustable persona controls. Publicly deployed on Streamlit.

🤝 YourTurn.group — group coordination, automated A Flask app that takes the grunt work out of running recurring group activities — scheduling, reminders, rotations, and communication — so the humans can focus on showing up. Built on Railway with Stripe, Postmark, Cloudflare R2, and Google Places.

Plus a production SaaS (PlanPull) that extracts material lists from contractor PDFs at 95%+ accuracy, an ML pipeline that predicts at-risk online students weeks before course end, and a pandemic-era SARIMAX forecasting model that hit <1% MAPE on 30-day infection-rate predictions. Browse the repos below for more.

Machine learning, LLM and agentic systems, evidence-based pedagogy for technical subjects, and equitable curriculum design. At Coding Dojo I led a team of six data science instructors, authored curriculum on NLP, LLMs, ML pipelines, model interpretability, and time-series forecasting, and maintained a 9/10 average student-feedback rating across cohorts. At LearningFuze I rebuilt a 12-week data science program into a 16-week program in four months as the sole instructor. I've also written articles on pretrained word embeddings, Keras callbacks, and other applied ML topics over at Medium / Towards Data Science.

🔭 Currently teaching generative AI to adult learners, consulting with small businesses and nonprofits, and applying for academic teaching positions where I can do this kind of work at scale.

I still love the outdoors and have traveled all over. Some of the most amazing people I've met have been in beautiful countries — and some of the most beautiful places have been right here in the United States. Have you been to Crater Lake in Oregon, or Yellowstone in Wyoming? You should.

How about the Golden Mountain in Ecuador, or Mt. Kenya in... Kenya? Travel is the best.

• 📧 Email: caellwyn@gmail.com
• ✍️ Articles: caellwyn.medium.com
• 💼 LinkedIn: linkedin.com/in/caellwyn
• 🌐 Consultancy: caellwynai.com

Or just scroll down and poke around my repositories.

Thanks for stopping by — have a great day!