This page prints as a two-page CV.

Steven Burns

AI solutions architect

I design and build systems where AI does real work, and I lead the teams that ship them. Twenty-five years across network engineering, web platforms and live experiential installations, now focused on LLM products, agent tooling and local-first generative pipelines.

Experience

Technical Owner, Enterprise AI learning platform (contract), Confidential client

2025–now

Turned a client’s open question, how do we use AI with our product, into a costed, phased plan and a six-week proof of concept on Bedrock, then took over their early-stage platform to bring the guide into it
Owned it end to end for 11 months: architecture, AI, infrastructure, design and interface, with about 73% of today’s code across ~1,170 commits
Grew it from one pilot to three signed enterprises, with a fourth close, each on its own isolated database and deployment, with content authored once and published read-only
Built the AI guide on Claude via Bedrock: layered prompts, long-term memory learners can read, retrieval and 18 tools in a streamed loop with guardrails
Tested it with agents: a synthetic cohort, a simulated working day, and a model-swap bench that found a higher rubric score (24/30 vs 19/30) at about 38% of the cost

Founder, AI Innovation and Solutions Architect, Cre8ive Industries LLC

2003–now

Fractional CTO for Cohezion, an AI startup: architected its AI infrastructure, mentored the dev team and led product strategy
Technical team manager for Arival, an international travel research firm: ran a distributed team and brought in agile process
Founded AI News Brew and ran it for three years: cross-sourced, bias-scored daily news with a haiku per story, closed when its data license ended with 5,738 articles archived
Generative media work licensed by Horiar to mature its platform; wrote about 20 MCP servers so agents render images, video, voice and music on local hardware
Built an LLM evaluation ladder that rated 44 local and cloud models over 378 matches, and distilled a 14B model that ranked fourth
Earlier: an on-demand movie service in 2003, e-commerce and CRM for ElfOnTheShelf.com, product search for FloorAndDecorOutlets.com, and an email platform that sent hundreds of thousands of emails

Director of Software Development, Toolbox No. 9

2022–2023

Defined the strategy for modernizing and maintaining legacy client applications
Supported business development through RFPs and proposals
Owned timelines, resourcing and quality standards across the portfolio

Director of Development and Innovation, LUCKIE

2021

Stood up a 24-hour development and delivery pipeline for a major pharmaceutical client
Managed local and international teams and tightened communication across them

VP of Development, Look Listen

2016–2021

Built technology for AT&T, Samsung, IHG, AIG, Chevron and LG
Mobile escape rooms, a Final Four basketball simulation and national STEM activations, on reusable frameworks of web, sockets and custom sensor boards
Managed about a dozen engineers and a director of development, including overseas contractors
Pitched and presented to clients, and spoke at industry events
Best in Show, AIMA 2018

Earlier career

2012–2014 Lead Web Developer, Manager of IT and Operations, Pinnacle Promotions
2005–2009 Manager of Technical Services, AS/AP Apparel Software
2002–2005 Senior Unix Systems Engineer, Glenayre Electronics
1998–2002 Network engineering and administration, Mentisys, Chatfish, Fong Kai Industrial

Selected work

Lull (2026). Calm lost my wife’s prepaid year and had no fix to offer, so I rebuilt what she actually used in a morning. Within two days it was a daily practice shared hand to hand: generated and recorded soundscapes, a breathing practice with streaks and push nudges, single-use invites, and a lock-screen mode that keeps sound playing overnight. No subscription. Live and invite-only, with the source on GitHub.
Enterprise AI learning platform (2025–2026). A client’s open question, how do we use AI with our product, became a costed plan, a six-week proof of concept, and then an enterprise learning platform I have owned end to end for 11 months. Its AI guide knows each learner and never invents what it doesn’t know. Three enterprises have signed, with a fourth close, each on its own isolated database. A model-swap bench found a higher rubric score at about 38% of the cost. The client is confidential.
Biome (2026). An evaluation lab built as a strategy game: 44 local and cloud models on a live ELO ladder, a deterministic ecosystem simulation judging every move, and self-play fine-tuning that produced a 14B model ranking fourth of 44, ahead of most cloud models on the same ladder.
Waywyrd (2026). A rune-practice app where the model has one narrow job: read each cast against researched rune, sky and stone lore, and never claim more than the sources support. A provider-agnostic gateway runs Claude on Bedrock alongside open-weight models via Ollama on the same sourced context. Friends invite friends, and a reading is never charged for.
Random Acts of Creativity with AI (2023–2026). Three years at the frontier of generative image, video and audio: hand-built ComfyUI experiments from 2023, work licensed by Horiar to mature its generative platform, and now a local pipeline. I have written about 20 MCP servers, including ComfyUI with 17 workflows, music and voice, so project agents render art, film, voice and music as part of the build in Lull, Waywyrd, Biome and Ninth Candle.
AI News Brew (2022–2025). My first launched AI product: I built the pipeline and set the editorial rules for three years of daily news, cross-sourced across dozens of outlets, bias-scored and distilled to a summary and a haiku, with a human approving what published. It closed when its data license ended; 5,738 articles remain as a searchable archive.
Experiential installations (2014–2021). Mobile escape rooms, a Final Four basketball simulation and STEM activations for national brands, built on web sockets, custom sensor boards and video capture. Best in Show, AIMA 2018.

Case studies: makeratl.com/work

Skills

AI and agents: Claude, OpenAI and Bedrock APIs, MCP servers and agent tooling, RAG and prompt caching, LLM evals and benchmarks, Ollama and local models, LoRA fine-tuning
Generative media: ComfyUI (Qwen Image, WAN 2.2, Hunyuan3D), ACE-Step music, Chatterbox voice, FFmpeg pipelines
Software: TypeScript and Node.js, Next.js and React, Python, PHP, WebSockets, Three.js
Data and infrastructure: PostgreSQL (Supabase, pgvector, RLS), MySQL, SQLite, Vector databases, AWS and Azure, Linux and Docker, High-availability architecture
Hardware: Arduino and custom PCBs, Sensors and IoT (AWS IoT), LED installations, 3D modeling and printing
Leadership: Building and mentoring teams, Distributed and overseas teams, Agile delivery, RFPs, pitches and client work
Certifications: AWS IoT, Red Hat Certified Engineer (RHCE), Red Hat Certified Technician (RHCT)