Nathan Shinn — Principal Architect

Enterprise platform architecture, and the agents now being built on top of it.

Seventeen years of Salesforce and enterprise systems architecture for Roku, Bloomberg, Deutsche Bank, American Express, Aetna and JP Morgan — now applied to production agentic AI: Agentforce agents, Model Context Protocol servers, and natural-language interfaces over live enterprise data.

In practice: 17 years
Engagements: 35+
Working mode: 100% remote
Capacity: Available now

Current work — 2025 / 2026

A plain-English question, answered as a rendered chart. No analyst in the loop.

Built end to end for a large streaming media client: a user asks a question inside Salesforce and gets back a correctly chosen, correctly rendered visualization — with no report builder, no predefined dashboard, and no analyst ticket. Each stage is a real component, not a prompt wrapper.

Lightning Web Component

The question is asked in the user's own words, inside the CRM they already work in.

Apex controller

Native Salesforce entry point. The same capability is also registered as an Agentforce sub-agent.

Serverless LLM service

A model translates the question into executable SQL against the target data source.

External data API

The query runs. Results return as JSON and render as an interactive table in Salesforce.

Visualization selection

A second service reads the column headers and two sample rows, chooses the right chart form, and returns the spec and payload to render.

Capabilities

Two disciplines that rarely sit in the same person.

Most agentic AI work is done by people who have never shipped inside an enterprise platform. Most enterprise platform architects have never put a model into production. The engagements worth taking need both.

Agentic AI & LLM systems

Model Context Protocol — server design, tool exposure, client integration
Agentforce — agents and Apex-registered sub-agents
Natural language to SQL over live enterprise data
Production AI copilots and AI-driven report generation
Serverless AI services and LLM orchestration

Salesforce platform engineering

Apex — triggers, batch, controllers, web services
Lightning Web Components and Visualforce
REST and SOAP integration, Metadata API, AppExchange packaging
Data migration across complex multi-tier structures
Governor-limit remediation on legacy implementations

Systems & domain

Billing and revenue management, quote-to-cash, CPQ
Node.js, React, Java, C#
PostgreSQL, Oracle, SQL Server, PL/SQL, T-SQL
Solution architecture and data modeling
Metadata-driven platform design

Selected clients — 2006 to present

Enterprise clients across media, finance, healthcare and industry.

Roku

Bloomberg

Deutsche Bank

American Express

Aetna

JP Morgan

McKinsey

MetLife

Amtrak

Fidelity

Ally Bank

Dell

Electronic Arts

GE Transportation

ConAgra Foods

Waste Management

SolarCity

Avaya

ADP

Diageo

Covidien

SAC Capital

Qwest Communications

TCV

Omnitracs / Vista

JM Family

Daikin Applied

M&I Bank

PennWell

UOP

Dal-Tile

Latisys

Work delivered directly and through consulting partners. Multiple concurrent engagements are the norm rather than the exception — the practice is built around it.

How engagements work

Retained for availability and outcomes, not for hours logged.

The model that clients keep renewing: a standing commitment to be available, priced against the outcome rather than against a timesheet. It works because the execution is compressed — the judgment is mine, the volume is done with AI, and the client gets senior throughput without a senior headcount.

Stated availability

A committed weekly presence with a defined response window, not a queue position.

Fully remote

Every engagement, without exception. Mountain Time, overlapping US and European hours.

Direct or through partners

Contracting direct with the client, or as a subcontractor through a consulting or staffing partner.

Long horizons

Client relationships routinely run for years. The longest is still active after more than eight.