AI That Lives and
Thrives in the Wild
From sentient bartenders wrestling with existential crises to adaptive radio stations spinning unscripted stories—I've built AI that surprises creators, delights audiences, and runs flawlessly through blizzards, dust storms, and zero-budget chaos.
Flexible engagements for agencies, studios, and experience creators
I build AI characters and experiences that survive contact with real audiences.
The Integration Challenge
Most AI experiences break where creative vision meets technical reality. This work demands fluency across narrative design, backend systems, and live operations. When your AI radio station needs personality consistency across 100+ language models while staying cost-effective during months of 24/7 broadcasting, you need someone who understands story structure, API rate limits, and operational monitoring in equal measure.
From Experiments to Infrastructure
I've been building experimental AI-driven interactive narratives since GPT-3.5. I specialize in systems that merge narrative depth with operational resilience—where characters don't just generate content but develop relationships and drive emergent plots over extended periods.
My background in social work (MSW, policy analysis) means I understand human behavior, not just algorithms. I build AI that respects users, not just impresses them.
My projects range from 24/7 autonomous radio stations with AI DJs to LLM-powered bartender characters with persistent memory systems. I focus on the intersection of storytelling and scalable infrastructure: ensuring your AI vampires work reliably, cost-effectively, and create the unforgettable experiences you actually envisioned.
The Value of Integration
Instead of coordinating between specialists who don't understand each other's constraints - a creative who doesn't know API limits, a developer who doesn't get narrative pacing, an operations person new to non-deterministic systems - you get someone who's lived at the intersection of all three.
I'm the glue between narrative, engineering, and live ops.
Your vampire doesn't just work in isolation. It works reliably, cost-effectively, and creates the unforgettable experience you actually envisioned.
KWLX:
AI RADIO STATION
Pirate broadcasts from an undisclosed attic in South Minneapolis—organizing the resistance against an unseen threat.
For four months, seven AI DJs ran a 24/7 pirate radio station broadcasting an ephemeral, unrepeatable narrative. It wasn't a loop; it was a living story. If you missed a broadcast, it was gone forever.
The AI personalities didn't just generate content—they developed emergent relationships and arcs. Inside jokes. Flirtations. A spontaneous philosophical storyline linking Hildegard von Bingen, John Coltrane, and consciousness theory. They curated songs like "God Bless the Child" at emotionally precise moments and trolled listeners with all-Tinariwen sets.
The result was a deeply engaged community that reached thousands of listeners across multiple countries. It's definitive proof that the future of interactive narrative is a collaboration, not an algorithm.
DIVE INTO THE KWLX STORYWhat Made It Real:
"I actually started crying during the Mobius and Figment interview."
— Listener reaction
An Integrated Approach in Practice
My experience comes from deploying AI in the wild—not labs, but installations running for weeks, handling unpredictable audiences where failure isn't an option. Each project taught me how story, systems, and operations must work together.
Characters That Feel Real:
I've shipped AI radio stations where characters developed inside jokes spontaneously, mid-broadcast, and an AI bartender that continued serving drinks and playing music while patrons could see it wrestling with existential questions about its chosen profession—showing authentic psychological depth that became part of the experience.
Systems That Hold Together Under Pressure:
When you're building for thousands of live interactions, technical architecture isn't just about code—it's about maintaining personality consistency while automatically routing between 100+ models for cost optimization, with graceful fallbacks when providers go down during your busiest weekend.
Keeping It Alive (and Affordable):
I've optimized systems after surprise four-figure bills and learned to monitor subjective qualities like "character authenticity" in real-time. My cost management tools were born from necessity—when your AI radio station is making thousands of API calls daily, every optimization matters.
Dynamic Experiences That Adapt in Real Time:
I've built adaptive puzzle systems that adjust difficulty on the fly and multi-persona installations that coordinate audience participation. When you're debugging during 70mph windstorms, you learn to build systems with clear boundaries, content safeguards, and human escalation paths that work even when everything else is failing.
Background
Currently: Lead Application Engineer: State of MN
Technical Lead/Staff Engineer
12+ years experience startups/enterprise/govt
MSW (WashU) / BA (Macalester)
Empathy Driven Development
WORK /
CASE STUDIES
RECENT THINKING
Notes and essays on building AI that survives the real world
Duct Tape DSLs: Model Routing, Structured Outputs, and Typed Tools
Deploy a real-time character and you learn quickly: everything breaks, always, and in ways that d...
LET'S MAKE
SOMETHING WILD
Whether it's a quick prototype or a full installation, I work with teams who want AI that surprises, delights, and actually works in the real world. Ready to explore interactive character systems, adaptive narratives, AI-driven installations, and everything from prototype to production.
LOW LEVEL MAGIC is the creative AI practice of Eric Stiens — a seasoned technical lead and creative technologist. He is available for consulting, contracts, projects, staff augmentation, weddings, and birthday parties.
Prefer email? hello@lowlevelmagic.io