decision
Decision systems
Systems that transform uncertain signals into actions, recommendations, plans, or institutional choices.
From uncertain signal to accountable action.
Decision systems / knowledge infrastructure / autonomy
Systems under real constraints.
I work with systems that make or support decisions under uncertainty. That has taken several forms across my career: quantitative models in live markets, enterprise machine-learning systems, generative AI research programs, autonomous agents, knowledge-discovery infrastructure, evaluation pipelines, and production architectures where reliability and trust matter.
My strongest work happens before a problem has been fully named: I help define the question, identify the hidden structure, build the first serious system, test the assumptions, and create the language and operating model needed for others to adopt what works. I am not interested in technology as theater; I am interested in systems that can survive contact with reality.
Scope protocolartifact first / no theater
question what needs to be understood, decided, or built
evidence what would make the answer trustworthy
artifact memo, prototype, review, map, or operating model
boundary privacy, authority, deadline, risk, and adoption path
response: concrete next step after scope
Selected systems work
Proof without a resume.
I have worked across several environments where models and software meet real consequence: quantitative systems exposed to live feedback, enterprise machine-learning programs, agentic knowledge infrastructure, and production autonomous systems.
The details differ. The pattern is consistent: uncertain signals, operating constraints, institutional risk, and the need for systems that can be evaluated, governed, and improved.
Selected traces: Relik AI, PwC AI Factory, production autonomous-agent systems, quantitative decision systems.
Work domains
Professional consequence.
The recurring surface is representation, memory, evaluation, and action: systems that help people and institutions understand more, decide better, and act with greater responsibility.
knowledge
Knowledge infrastructure
Tools for memory, retrieval, provenance, synthesis, source-grounded discovery, and the preservation of context.
Memory, retrieval, and synthesis without losing the source.
autonomy
Accountable autonomy
Autonomous and semi-autonomous systems that remain observable, bounded, evaluable, corrigible, and aligned with legitimate human authority.
Autonomy with visible limits and real oversight.
institution
Institutional cognition
Human-machine systems that help organizations form better beliefs, allocate attention, make decisions, and learn from outcomes.
Organizations learning through better tools and feedback.
governance
Evaluation and governance
The practical control surfaces that determine whether powerful systems can be trusted: data, memory, feedback, tests, escalation, observability, permissions, failure detection, and human oversight.
Capability made legible, reliable, and governable.
Engagement modes
Ways work can happen.
The form stays secondary to the problem. A useful engagement produces a decision, a working surface, a map, or an operating model that can be tested.
mode
use when
fit
output
consult
Strategic judgment
A focused working session for technical strategy, product architecture, institutional AI, or a consequential decision under uncertainty.
fit
Best when the expensive part is deciding what is true, possible, or worth doing.
strategy / architecture / judgment
output
Scoped call / written follow-up
Decision memo
first packet
Send the decision, the audience, the deadline, and what the answer needs to settle.
first packetSend the decision, the audience, the deadline, and what the answer needs to settle.
build
Build sprint
A short implementation push for prototypes, internal tools, evaluation harnesses, knowledge workflows, or agentic systems.
fit
Best when the useful thing is a working surface, script, prototype, or harness.
prototype / harness / workflow
output
Defined deliverable / direct execution
Usable system
first packet
Send the current system, intended user, success test, and what can safely be ignored.
first packetSend the current system, intended user, success test, and what can safely be ignored.
brief
Research brief
A compact synthesis of a technical area, market shift, model capability, institutional risk, or design space with recommendations.
fit
Best when scattered evidence needs to become a clear map and next move.
synthesis / map / recommendation
output
Memo / sources / next steps
Research packet
first packet
Send the question, sources you trust, sources you distrust, and when the map is needed.
first packetSend the question, sources you trust, sources you distrust, and when the map is needed.
Scope practice
How I scope.
We start with the decision, constraint, or artifact that would make the work valuable. From there, the engagement can stay small or expand only when the evidence supports it.
Scope before commitment
The first useful output is a clean shape of the problem.
Artifacts over theater
Calls are useful only when they make the work sharper.
Defaults with tradeoffs
I will pick a path, name the costs, and show the evidence.
Private stays private
Sensitive work is handled with explicit boundaries.
First note
faster scope / fewer loopscontext
What exists now, what changed, and who the work is for.
constraint
What cannot move: deadline, stack, audience, budget, risk, or privacy boundary.
artifact
The concrete thing that should exist afterward: memo, prototype, review, map, or plan.
If the work involves sensitive strategy, unreleased systems, or private data, send only the outline first. The secure channel comes after scope is clear.
private reply after scope