I build interactive AI products, persistent world systems, and multimodal storytelling pipelines across the full stack. I’m an active developer on Marinara Engine and a computer science student at the University of Colorado Boulder.
Open to AI engineering, full-stack software, and interactive-media opportunities.
Role: AI Systems Architect & Open-Source Developer
Stack: TypeScript · React · Node.js · pnpm · local AI · multimodal generation · persistent world state
Marinara is a local-first AI chat, roleplay, and game engine. I’ve authored 100+ merged pull requests spanning client and server architecture, shared contracts, transactional persistence, downloadable capability APIs, prompt orchestration, multimodal media, responsive UI, regression testing, and documentation.
- Hierarchical world state & spatial context — built persistent nested locations, graph validation, atomic/idempotent movement, snapshots/checkpoints, bounded prompt projection, AI map drafting/expansion, lore grounding, and responsive authoring UI.
- Capability API 1.3 host services — exposed safe package access to model routing, resources, persistence, transactions, history/checkpoints, logging, and client contribution lifecycle without leaking private Engine internals.
- Storyboards for Roleplay — extended multimodal storytelling beyond Game Mode with episode cadence, inline still/video playback, provider-neutral motion planning, prompt libraries, and per-chat generation controls.
- Shared World Maps artwork — added account-owned visual references across chats with guarded deletion, transactional lifecycle locks, race-safe reference creation, and shared-world integration.
- Unified travel & recovery — implemented server-owned route validation, step-by-step/instant travel, one canonical movement mutation per accepted turn, command-ID recovery, retry preservation, and deterministic spatial regressions.
- Hidden directive stream/persistence hardening — sanitized model-only movement directives across streaming, rewriting, impersonation, and persistence while preserving authoritative state transitions.
View all Marinara pull requests →
- AI systems: LLM integrations, agent/runtime architecture, prompt composition, image/video generation, provider-aware pipelines
- State & reliability: transactions, idempotency, revision conflicts, race handling, snapshots/checkpoints, recovery paths
- Full stack: TypeScript, React, Node.js, REST, WebSockets, React Query, responsive/PWA interfaces
- Extensibility: versioned capability contracts, package lifecycle/readiness, host APIs, shared resource boundaries
- Validation: deterministic regressions, Playwright, lint/type/build gates, container validation, browser QA
Voice-driven Socratic learning across web and VR, built with React, TypeScript, Supabase/Postgres, Node.js, real-time voice AI, and Unity XR prototypes.
Python and GitHub Actions pipeline that ingests RSS feeds, filters and summarizes AI research/news with LLMs, and publishes a concise daily brief.
Contributed to an AI/VR prototype involving assistant workflows, rover-autonomy concepts, and AR overlays.
- University of Colorado Boulder — B.A. Computer Science, expected May 2027
- Front Range Community College — A.S. Computer Science, 2025
- NVIDIA-Certified Associate — AI Infrastructure & Operations, 2025