AI Integration · Backend Systems · Kotlin · TypeScript · Python · Rust

John Noecker Jr.
I ship products with AI built in.

Product-oriented software engineer with 15+ years shipping complete systems end-to-end. Recent work focuses on integrating LLMs, generative AI, and ML pipelines into production applications — from multi-provider AI generation platforms to computational linguistics toolkits with neural embeddings. Deep backend expertise with full-stack fluency across Kotlin, TypeScript, Python, and Rust.

Email me Resume (PDF)

165K+ lines across 3 production systems
2,000+ tests — deterministic, coroutine, integration
5 AI image providers · 3 LLM backends · NLP pipelines
Tauri 2 · React 19 · FastAPI · Ktor · PixiJS
Published on PyPI · 14 academic papers · 2 U.S. patents

Recent projects

Arcanum (opens in new tab)

100K+ line desktop worldbuilding tool and cloud publishing platform for multiplayer game creators. Tauri 2 desktop app with 291 React components, 104 async Rust backend commands, and a 69-panel editor — with AI-assisted content creation woven throughout via five image providers, three LLM backends, and a text-to-speech pipeline.

TypeScript Rust React 19 Tauri 2 Zustand Claude API OpenAI Runware FLUX DeepInfra Cloudflare Workers D1 R2

Unified AI dispatcher integrating Runware FLUX, OpenAI GPT Image, and DeepInfra for image generation plus Claude, DeepSeek, and OpenAI for text — with automatic prompt enhancement, style templates, and background removal
Serverless cloud platform (Workers + D1 + R2) with AI proxy layer enforcing model allowlisting and per-user lifetime quotas; wildcard-subdomain publishing deploys each world as an independent showcase SPA
AI-powered lore authoring suite: gap analysis, consistency auditing, vision-based map analysis, relationship discovery, timeline inference, and cinematic video export via FFmpeg
Format-preserving YAML round-trip I/O retains comments and key ordering; content-addressed SHA-256 asset storage with lossy WebP re-encoding for publish

AmbonMUD (opens in new tab) Live demo (opens in new tab)

50K+ line production multiplayer game server built using AI-assisted development workflows (Claude Code, Codex). Coroutine-based tick engine, 10+ interconnected game systems, 196 command variants, and a React/PixiJS web client — architected for horizontal scaling from standalone to sharded deployment.

Kotlin Coroutines Ktor gRPC PostgreSQL Redis React 19 PixiJS 8 Prometheus Docker

Single-threaded coroutine engine eliminates concurrency bugs by design; tick-based architecture processes all game state transitions deterministically
Horizontal scaling: standalone mode handles hundreds of concurrent users; zone-based sharding over gRPC with Redis-backed event coordination supports near-unlimited capacity across container instances
10+ game systems — real-time combat with threat tables, behavior-tree NPC AI, procedural dungeons, crafting/gathering economy, guild halls, housing, quests, achievements, pets, and PvP
React/PixiJS web client with canvas-rendered battle scenes, structured GMCP protocol, and full UI; 500+ tests with deterministic time control and CI/CD via GitHub Actions

Mowen (墨紋) (opens in new tab)

Full-stack authorship attribution toolkit — clean-room Python successor to JGAAP (opens in new tab), the system used to identify J.K. Rowling as Robert Galbraith. 16K+ lines with 113 pluggable pipeline components, dual execution paths (traditional stylometry + neural embeddings), and 790+ tests. Published on PyPI as three packages.

Python FastAPI SQLAlchemy HuggingFace spaCy React 19 Zustand TanStack Query Docker PyPI

5-stage NLP pipeline (Canonicize → Extract → Cull → Histogram → Analyze) with 41 feature extractors, 26 distance functions, 21 classifiers, and a generic type-safe Registry[T] pattern — one file and one decorator to add a component
Dual execution paths: discrete stylometry (n-grams, POS tags, function words across 10 languages) and neural embeddings (HuggingFace transformers, SELMA instruction-tuned models, GNN over dependency parses, GPT-2 perplexity)
Authorship verification via General Imposters Method and Unmasking with dual-threshold calibration; style-change detection at paragraph boundaries; evaluation across EER, c@1, AUROC, and Brier score
Zero-dependency core with 12 optional extras gated behind lazy imports; FastAPI backend with background experiment execution; React frontend with real-time progress tracking and client-side metric computation

Stack

Tools and technologies across recent project work.

Languages

Kotlin, Java, TypeScript, Python, Rust, SQL

AI & ML

LLM integration (Claude, OpenAI, DeepSeek/OpenRouter), image generation (Runware FLUX, GPT Image, DeepInfra), HuggingFace Transformers, embeddings, NLP pipelines, prompt engineering, text-to-speech

Backend

Coroutines, Ktor, FastAPI, Spring Boot, gRPC/Protobuf, Node.js, Jackson

Frontend

React 19, Tauri 2, PixiJS 8, Zustand, XY Flow, TipTap, Leaflet, Recharts, Vite, Tailwind CSS

Data

PostgreSQL, Redis, Cloudflare D1, SQLAlchemy, Exposed, HikariCP, Flyway

Infrastructure

Docker, GitHub Actions, Cloudflare Workers/R2, Prometheus, Grafana, CI/CD

Quality

JUnit, pytest, Vitest, deterministic testing, coroutine testing, integration suites

Timeline

Where I've been, and the kinds of problems I've owned.

Noecker & Associates

Staff Software Engineer (Independent) · Aug 2020 — Present

Solo-designed and shipped three production applications with AI integration throughout — a desktop worldbuilding platform, an NLP toolkit, and a multiplayer game server — 165K+ lines across Kotlin, TypeScript, Python, and Rust.
Built unified AI generation pipelines dispatching across five image providers and three LLM backends with prompt enhancement, style templates, and serverless AI proxying with per-user quota enforcement.
Implemented dual-path NLP analysis combining traditional stylometric features with HuggingFace transformer embeddings, instruction-tuned models, and graph neural networks.
Delivered all projects with comprehensive testing (2,000+ tests combined), CI/CD, Docker deployment, and production observability.

Enzyme

Senior Software Engineer · Jan 2019 — Jul 2020

Greenfield platform (TypeScript, React, Node.js) for FDA 510(k) regulatory data extraction.
Backend document processing with strong correctness, auditability, and compliance requirements.

Signifyd

Software Engineer · Aug 2016 — Dec 2018

~$100K/month recurring data cost reduction via vendor evaluation and integration.
Backend throughput and latency improvements via caching and performance tuning at scale.

Juola & Associates

Software Engineer · Aug 2010 — Jul 2016

Forensic authorship attribution systems for national security and litigation; methodologies passed Daubert admissibility challenges.
Peer-reviewed research, $1.6M grant contribution, and two U.S. patents co-authored.

EVLL (Applied Research)

Software Engineer · Aug 2010 — Oct 2014

Computational stylometry for high-profile cases including The Cuckoo's Calling (J.K. Rowling) and the Satoshi Nakamoto investigation.
Co-developed JGAAP, the open-source authorship attribution framework later succeeded by Mowen.

PROTEUS Technologies

Software Engineer · May 2009 — Aug 2010