Useful agents.
Accountable work.

Start with work people already do

Enterprise AI agents help people complete a defined job using company knowledge and approved tools. My work spans project context, decision support, executive communication and a pilot for test-fitting support. I also design more complex workflows whose deployment is still ahead.

My background and Berkeley education

I completed Agentic AI: Strategy, Applications and Organizational Impact through UC Berkeley Executive Education, Haas. I also hold AWS Certified AI Practitioner. I apply that education alongside hands-on product work: connect the business problem, the technical design and the people accountable for the result.

How I approach enterprise agents

I start with a painful workflow and map its steps, sources, decisions and owners. I then give the agent a narrow job, ground it in authoritative knowledge, use deterministic rules for calculations, and make approval and escalation explicit. I test real scenarios, pilot with users and expand only when the evidence supports the next step.

The progression is find information, synthesize context, reason about options, assist a person, execute a bounded step and eventually coordinate several workflows. Human approval belongs at consequential decisions throughout that progression.

Check N. Herder: project context

Stage: Created and shared.

Problem: Project decisions disappear into chats, meeting notes and scattered updates, leaving people to reconstruct what happened.

Solution: I created Check N. Herder to turn that history into a shared picture of the problem, plan, progress, owners and next actions.

Technical solution: The Teams-centered workflow reconstructs context, proposes source-backed baselines, tracks changes in scope and priorities, and prepares red, yellow or green status updates with risks and decisions.

Outcome: The agent has been created and shared; faster handoffs and less reporting work are its intended value, with no measured time savings claimed here.

Council Decision Lab: compare the options

Stage: Created and shared.

Problem: A difficult business decision can look very different to a researcher, a product manager and an architect.

Solution: I created Council Decision Lab to examine the same question through several perspectives before forming a recommendation.

Technical solution: Its multi-persona reasoning pattern combines research, executive, data, architecture, product and program perspectives with challenge and audit passes, then synthesizes evidence, decision criteria, trade-offs, risks and an execution plan.

Outcome: The agent has been created and shared; the result is a structured decision aid for a human decision-maker, not a claim that a council guarantees a correct decision.

Company deck builder: explain the decision

Stage: Created.

Problem: Technical and project information often needs substantial rewriting before leaders can use it to make a decision.

Solution: I created a company deck builder to turn fragmented material into a clear executive story.

Technical solution: The workflow organizes source material, synthesizes the evidence, frames it for the audience and produces presentation-ready narratives, business cases and status content.

Outcome: The agent has been created; it produces drafts for review, while reductions in slide preparation time remain unmeasured.

Test-fitting support and intake agent

Stage: Published internally; pilot.

Problem: Employees need help using a test-fitting tool, but onboarding questions and requests can get lost across chats and email.

Solution: I built an internal support agent to answer approved how-to questions and capture onboarding, access, support and enhancement requests.

Technical solution: Microsoft Copilot Studio, Teams and Microsoft 365 connect approved knowledge and training to classified SharePoint intake and notifications; requests require confirmation and a person reviews them before the external vendor process.

Outcome: The agent is published internally, has been evaluated and is being piloted; it supports administration and routing and does not make engineering design recommendations.

Test-fitting workflow agent

Stage: Specified; deployment not established.

Problem: An early site concept requires complete inputs, site constraints, suitable product templates and a sequence of design checks.

Solution: I specified a human-supervised agent to help a design manager move a concept-fit request through the test-fitting workflow.

Technical solution: The design includes a machine-readable SOP, intake fields, atomic steps, a constraint model, responsibility mapping, tool actions, quality checks, source traceability, exception handling and approval gates.

Outcome: The requirements and architecture are specified; automated layout generation, historical replay and production execution remain intended capabilities rather than deployed results.

Development intelligence: connect the work

Stage: Product architecture; proof-of-concept stage.

Problem: Customer requirements must be translated repeatedly into site, product, design, cost, supply and delivery decisions.

Solution: I designed a development intelligence layer to carry a shared set of assumptions and evidence across those decisions.

Technical solution: The architecture consumes specialist systems and coordinates source, customer, product, site, sizing, test-fitting, cost, procurement and decision workflows, using deterministic calculations and human approvals where appropriate.

Outcome: This is product architecture and proof-of-concept work; it is not a completed end-to-end autonomous platform and has no claimed production savings.

What I measure before expanding

I look for correct task completion, source coverage, safe tool use, fewer unnecessary handoffs and the amount of review still required. A useful pilot establishes a baseline before anyone attaches a savings claim to it. These examples separate created agents, an internal pilot and specified systems so the stage is clear.