AI Engineering Portfolio
I'm Spencer Marks — an engineering leader who build things.
These applications are working AI-powered systems I've designed, built, and deployed. They run on local infrastructure using open-source LLMs with zero cloud dependencies, demonstrating privacy-first AI architecture suitable for regulated environments. Each project is built using AI-augmented development workflows with Claude Code.
Stack: Python, Django, FastAPI, Ollama (Qwen 32B), ChromaDB, SwiftUI/SwiftData, AWS, PostgreSQL
AI tooling: Claude Code with custom CLAUDE.md project context, RAG pipelines, vector embeddings, LLM context injection
Practices: Pre-commit CI (ruff, pyright, pytest), automated coverage gates, git worktrees, justfile automation
AI tooling: Claude Code with custom CLAUDE.md project context, RAG pipelines, vector embeddings, LLM context injection
Practices: Pre-commit CI (ruff, pyright, pytest), automated coverage gates, git worktrees, justfile automation
Tarmar
Full-stack AI + web platform for tabletop RPG. AI-powered game master with LLM context injection from structured rulesets, plus Django character management application.
Ollama, Qwen 32B, Django, Python, pre-commit CI, Beads task tracking
tarmar.origamisoftware.com
Photo Catalog Assistant
RAG-powered natural language search across terabytes of photography metadata. Ask questions like "show me golden hour shots from 2019" and get structured results.
ChromaDB, Ollama embeddings, exiftool, Python
photos.origamisoftware.com
Virtual Card Table
Digital card game table for multiplayer tabletop gaming.
cards.origamisoftware.com
Non-AI Applications
These tools don't use AI at runtime — they're conventional software. I built them with Claude Code as my pair programmer, the same AI-augmented workflow as the projects above.
Watcher
Website change monitor — polls one URL or many, detects content changes via SHA-256 hashing, and sends SMS alerts through ClickSend. Supports continuous background runs with randomized intervals and per-URL retry/backoff.
Python, ClickSend SMS
iSpend (iOS app)
A mindful-spending tracker for iPhone, iPad, and Mac. Intentionally manual — no bank sync, no auto-categorizer, no AI guessing on your behalf. You enter every expense yourself, and that small moment of friction is the whole point. Borrows from cognitive-behavioral techniques to interrupt impulse spending and make values-aligned choices easier.
SwiftUI, SwiftData, CloudKit
App Store — coming soon
If you'd like to try any of these applications, I'm happy to set up access. Tarmar has both public and registered modes.