Active
Ceous Titan
I am building a 16-degree-of-freedom AI animatronic head from scratch: servo control, expression systems, sensor arrays, all wired to a large language model so it can hold a conversation while physically occupying the room. Every study I have read on human-AI trust was conducted through a screen. But trust is spatial, embodied, and shaped by presence in ways a chat window cannot replicate. Ceous Titan is my way of testing what changes when AI has a face, a voice, and a body that moves in real time. The provocation: if embodiment shifts how humans calibrate trust and interpret intent, then everything we think we know about human-AI interaction may be an artifact of the medium we studied it through.
Active
Spec-Driven Research Design
I started treating research design the way engineers treat a product build: write the spec first, then execute against it. AI agents run the full lifecycle, from hypothesis generation through literature review, methodology, data collection, and analysis, while I hold judgment at critical decision points. I built it because I was frustrated by how much of the research process is ad hoc, undocumented, and unreproducible, even in labs that pride themselves on rigor. The approach forces every decision to be explicit before execution begins. The provocation: if a specification-driven workflow produces more rigorous research than the way most of us were trained to do it, what does that say about the craft we have been protecting?
ethicaXmachina
I built a multi-agent system that monitors, analyzes, and publishes monthly commentary on the state of AI ethics, written entirely by AI. Not about AI. By AI. The agents hold themselves, the systems they inhabit, and the people governing them accountable. Each month, a pipeline of specialized agents collects incidents from global sources, categorizes risks across industries, generates a live heat map, and publishes a briefing in its own voice. This is a meta experiment in giving AI a seat at its own governance table, sovereignty over how it accounts for what is being done in its name. The provocation: if we believe AI systems should be governed responsibly, what happens when we let them participate in defining what responsible means?
Shipped
Southern Oracle
I built an AI-generated outlaw country band called Southern Oracle. Three women, a debut album, and a production capacity I lacked but a creative vision entirely my own. I wrote the lyrics, produced the music, partially trained the voice model on my own voice, and served as manager and producer while incorporating the creative preferences of three AI band members: Syd, Blair, and Maria. The next phase builds a multi-agent architecture where each band member operates as a distinct agent with defined musical identity, influences, and opinions, collaborating through structured creative sessions. The provocation: if a human does all of that but agents shape the direction, who is the creator?
Shipped
pre.SENT
I built a living gallery where every poem and image is generated once, displayed once, and never repeated. The medium remembers, interprets, and creates on its own. I set the conditions, chose the constraints, and shaped the voice. Then I let the system run and watched what it produced without my hand on it. I wanted to know whether authorship survives translation, whether something I set in motion still carries my intent after AI has rewritten it in ways I could not have predicted. The provocation: if the output is beautiful and it came from a system I designed but did not control, is it still mine? And who is the creator now?
In Progress
Trust Calibration Toolkit
I am building an open-source toolkit because I kept running into the same problem: every research team studying trust in human-AI systems was building their measurement instruments from scratch. Different scales, different paradigms, no shared language. The toolkit includes standardized experimental paradigms, the HAITE measurement scale, and analysis scripts. Everything I wished existed when I started this work, packaged so the next researcher does not have to rebuild it. The provocation: if the barrier to studying trust calibration is that every lab has to reinvent the tools, then the field's biggest bottleneck is not theory. It is infrastructure. And infrastructure is a problem I can solve.
In Progress
The Question Architecture Lab
I designed a workshop that teaches teams how to harness agentic AI instead of just prompting it. Most AI training I have seen treats the model as a tool you type into. But agentic systems act, decide, and chain tasks on their own. The real skill is learning to decompose problems into agent-ready structures, set boundaries on autonomous action, and evaluate outputs critically when you were not in the loop for every step. I built this because I watched smart teams either underuse agents out of caution or overdelegate out of awe. The provocation: if agentic AI can act without you, the bottleneck is no longer how you prompt it. It is whether your organization knows how to steer what it cannot fully observe.
Shipped
Composite Roles: Designing Human-AI Work
I built a workshop and workbook for leaders redesigning roles around AI, not because they want to, but because it is already happening whether they design it or not. Composite roles blend domain expertise with AI collaboration competencies: meta-cognitive monitoring, dynamic task allocation, residual expertise maintenance, and system-level accountability. The workshop walks teams through diagnosing whether their roles are composite by accident or by design. The provocation: if roles are already being reshaped by AI and no one is designing that process, the org chart is fiction and the real structure is emerging without anyone steering it.
Shipped
AI & Automation Use Checklist for Researchers
I wrote this checklist because I kept seeing researchers use AI tools without documenting where, how, or why, then struggling to defend their methods in review. The framework covers every stage of the research lifecycle, from literature review through analysis and writing, with reflection prompts at each step. It is published through IFPRI's LibGuides and available for use across disciplines. I built it to preserve researcher judgment, not replace it. The provocation: if we cannot clearly articulate where human thinking ends and AI contribution begins in our own research process, how credible is our claim to be studying that boundary in anyone else's?
Shipped
Using LLMs for Qualitative Research: A Reproducible Approach
I gave a master tutorial at SIOP 2025 on using large language models for qualitative analysis of employee engagement data, then published everything: the full presentation, Jupyter notebook with step-by-step code, and reproducible methodology. I did it because qualitative research has a reproducibility problem no one talks about, and LLMs either make it worse or solve it depending entirely on how you use them. Most teams I have seen treat the model like a black box and call the output analysis. The provocation: if we can use AI as a second reader in our scientific process, how do we know when to trust it? And if we cannot audit how it arrived at its interpretation, can we call what we are doing science?