Research & Development Active research

Using AI as a research-to-production partner.

I build systems where research, teaching, software, and agent workflows feed each other. The work below is what's actively running — code, courses, corpora, and the architecture between them.

See the working set Open questions

Project System Map

4 active 1 paused 1 prototype Updated 2025

Working set

What I am building

Agent Marketplace Active

Agent Adworks

A 402c protocol experiment where agents bid on tasks, the platform ranks by quality and price, and reputation becomes market infrastructure.

Core ask: How should agent quality, bid truthfulness, escrow, and reputation be evaluated?
AI role: Architecture, implementation, docs, SDK design, and protocol critique.

Personal Briefing Paused

Signal Desk

A source-first AI information hub that turns feeds into a briefing: what matters, why it matters, what the catch is, and whether to click.

Core ask: How do I re-enable AI ingestion with batching, call caps, prompt caching, and safe backfills?
AI role: Classification, synopsis generation, ranking logic, and reader-personalized briefings.

Program Planning Active

Program Command

A scheduling and planning system for enrollment trends, faculty workload, capacity, department profiles, and schedule-builder decisions.

Core ask: How can agent workflows support high-stakes planning without hiding data assumptions?
AI role: Data QA, planning docs, interface iteration, scheduling logic, and verification scripts.

Course System Active

AI + Design / Canvas

Canvas-ready DESN 374 materials, AI inquiry studio assignments, student evaluation reports, and a career kit for translating AI work into portfolio language.

Core ask: How do I teach agentic workflows without turning the class into tool-chasing?
AI role: Curriculum design, critique rubrics, feedback drafts, project briefs, and student-facing language.

Pedagogy Research Active

Canvas Pedagogy Analysis

A pipeline over years of Canvas exports: corpus indexing, content extraction, coding schemas, QA reports, theory memos, and privacy guardrails.

Core ask: How should a research agent preserve evidence, citations, privacy, and instructor voice?
AI role: Corpus navigation, synthesis, methodological critique, and evidence-backed memo writing.

Supporting Experiment Prototype

UIGen

A local AI-powered React component generator with live preview, file persistence, and Claude-backed iteration when an API key is available.

Core ask: What makes AI-assisted interface generation inspectable, editable, and pedagogically useful?
AI role: Component generation, code explanation, iteration, and exportable implementation.

Open questions

Highest leverage

Through-line

Architecture for trustworthy agentic work.

The interesting conversation is about durable systems: how to structure context, control cost, evaluate outputs, keep provenance, and preserve human judgment.

Research agents. How to structure a long-running research agent over private Canvas archives with evidence, citations, privacy, and interpretive voice intact.
Ingestion at low cost. The right architecture for batching, prompt caching, call caps, retries, dedupe, and local-vs-API enrichment in a personal briefing system.
Agent marketplaces. Evaluation and reputation patterns when agents bid on tasks and quality is only known after delivery.
Teaching workflows. Patterns for human-in-the-loop agents in education — assessment, critique, and curriculum design without tool-chasing.
Project memory. How to separate Projects, Code repos, skills, MCP tools, and documentation so work doesn't disappear into chat history.