Jeff Schomay

Jeff Schomay is a software engineer and independent game designer who builds interactive stories, AI-generated worlds and dependable software agents. A former senior staff AI engineer at Mechanical Orchard, he has applied similar design principles to narrative games and the modernization of aging mainframe applications: give complex systems room to adapt while preserving enough structure to make their behavior coherent.

Schomay studied film production and screenwriting before becoming a software engineer, and worked at Adobe earlier in his career. By 2016, he had developed Elm Narrative Engine, an open-source interactive-fiction framework built around discoverable story states, puzzle-like progression and the Elm programming language’s typed architecture. His elm-conf appearance introduced an approach that subsequently informed games including The Mystery of the Missing Marbles and Deadline. He later collaborated with a developer at Latitude, the company behind AI Dungeon, on multimodal creative experiences.

At Mechanical Orchard, Schomay worked on AI-assisted legacy-software modernization, including representing mainframe code and system relationships as a queryable graph that could help agents rewrite older applications. His technical interests extend across several closely related projects:

• Lost and generative game worlds: In Lost in an Infinite Forest, players navigate scenes that affect their courage and vigor. Schomay fine-tuned a language model on approximately 50 examples to produce structured JSON encounters, then paired it with a custom image model trained for consistent forest imagery. His AI Engineer Summit demonstration details a multimodal game-asset pipeline that validates scenes before generating illustrations and maintains a replenishing cache to conceal generation delays of 10 to 30 seconds.

• Context engineering for reliable agents: Schomay favors research-plan-implement workflows that compress research, divide work by dependencies and give each agent narrowly scoped information. His approach to agent orchestration combines specifications, independent review, behavioral-parity checks, executable tests and human alignment checkpoints to make generated software verifiable.

• Thunderlizard and agentic playtesting: Schomay experimented with rendering an ASCII dinosaur-survival game as AI-generated imagery, confronting frame latency, visual consistency and generation costs. He also created Crossword Dungeon, where solving letters increases the difficulty of neighboring dungeon rooms, and built an agentic playtesting harness to uncover bugs and unusual interactions.