Zach Harris

BUILDING AT THE EDGE

linkedin.com/in/zharris · github.com/flashesofbrilliance

BUILDING AT THE EDGE · APPLIED AI · DECISION INTELLIGENCE

Creative technologist and builder with two decades across tech, marketing, data, and commercial operations in sports, finance, health, and consumer. Builds the intelligence layer before the category for it exists: structured pre-decision framing, AI-native analysis pipelines, and feedback loops that close each decision against its recorded outcome. Currently: ARCS, a decision intelligence platform built on Claude.

EXPERIENCE

ARCS

2025→

GRAYSCALE

2024–25

CONSULTING ↓

2019–23

CTF

2018–19

WTA

2016–18

◈ARCS · DECISION INTELLIGENCE PLATFORM

2025–Present

Built a decision intelligence platform from scratch: structured pre/post-decision capture, AI-native analysis pipeline (Claude, MCP), and a scoring loop that closes each decision against its recorded outcome. Built solo and in the open: 5,300+ commits in 195 days.

1,580+

PRS MERGED, EACH HUMAN-APPROVED

Shipped solo and in the open: 5,300+ commits in 195 days (as of Oct 21, 2026), every figure on status.arcs.care measured from the repository itself

Built AI-native analysis pipeline on Claude; MCP server + agentic orchestration layer handles pre/post-decision capture automatically

Designed the outcome-capture method: each decision is scored for confidence when made and closed against what actually happened; no accuracy figure is published until enough outcomes close

Platform guardrails as code: per-tenant Postgres roles with row-level security that fail closed, generated route and tool inventories with drift checks, and pre-commit gates that block whole classes of bugs. Live app at app.arcs.care; commit history and traces on request

Human-in-the-loop dev harness: parallel agent sessions, merges pinned to the approved commit, an append-only decision ledger in Postgres. 1,580+ PRs merged, 99 in one day, a human approving every merge. Live counts: status.arcs.care; release notes: one day, 99 merges

Compounding by design: every merge is receipted to a generated tool graph whose outcome weights steer the next build; graph retrieval beat the flat catalog 16/18 to 6/18 in a pre-registered blind test

Operating the full stack: go-to-market, growth motion, product development, commercial strategy. Breadth is the product

COMPASS SIGNAL

"What minimum viable intelligence system does a multi-BU platform need to compound learning across business units, and what's the right first wedge to prove the model?"

▣GRAYSCALE INVESTMENTS (DCG)

2024–2025

Led go-to-market for 10 SEC/FINRA-regulated digital-asset investment products ($7B AUM): positioning, segmentation, and adoption strategy inside regulated controls and governance.

$7B

AUM · REGULATED PRODUCT SUITE

Led go-to-market and market-expansion strategy across 10 regulated digital-asset products, inside SEC/FINRA controls and governance

Owned positioning, segmentation, and adoption strategy across the investor experience

Coordinated across product, engineering, legal, risk, and compliance through regulated release gates

Translated regulatory requirements into product and go-to-market requirements

"How do you take ten regulated investment products to market inside SEC/FINRA controls, and what does the governance-first go-to-market sequence look like?"

◇CHILDREN'S TUMOR FOUNDATION

2018–2019

Enterprise data and decision-intelligence layer from scratch: CDP architecture, CRM modernization, and segmentation across marketing, fundraising, and clinical. 20% net-new revenue growth through data-driven lifecycle strategy.

20%

NET-NEW REVENUE VIA DATA PERSONALIZATION

Built the enterprise data infrastructure end-to-end: CDP evaluation, CRM modernization (Salesforce Lightning), lifecycle segmentation

Built segmentation engine and scalable data model supporting marketing, fundraising, and clinical operations across 3 business units; the first intelligence layer any of these teams had

Drove 20% increase in net-new revenue through data-driven personalization, behavioral segmentation, and cross-BU growth instrumentation

"When a mission-driven org needs to shift from intuition-led to data-led fundraising, what's the right sequence, and how do you get program teams to act on the signal before the next campaign launches?"

△WOMEN'S TENNIS ASSOCIATION

2016–2018

Built the first fan behavior intelligence layer for a global sports platform: real-time scoring at <120ms, CDP architecture, and $850K+ in year-one commercial revenue.

<120ms

REAL-TIME FAN BEHAVIOR SCORING

Led CDP evaluation and selection for global consumer sports platform; built real-time fan behavior scoring at <120ms, serving personalized offers during live events

Designed first commercial partnerships with Twitter and Facebook: behavioral data flows, attribution models, event-driven revenue; $850K+ year-one

Pre/post-event intelligence loop: pre-tournament fan segmentation → in-event scoring → post-event outcome capture. Applied before "growth intelligence" was a category.

Platform load simulation and reliability engineering ahead of Women's Finals: data-informed capacity decisions, not guesses

"How do you build commercial intelligence for live events when the decision window is under 120 seconds and behavioral signal arrives asynchronously, and what does the pre/post-event capture loop look like?"

SELECT CONSULTING WORK

⬡AXON ENTERPRISES

GovTech · Jun–Dec 2023

New venture incubation: product instrumentation framework and KPI taxonomy for secure first-responder OTT content platform MVP.

MVP

SECURE OTT PLATFORM

Led discovery and business model validation for First Responder OTT network

Defined MVP features, instrumentation requirements, and success metrics for secure, video-first content platform

◑TWILL HEALTH

HealthTech · Jan 2021–May 2023

Market viability testing for 3 AI-enabled digital health product lines across a HIPAA-regulated environment.

AI HEALTH PRODUCT LINES VALIDATED

Conducted market viability testing for 3 AI-enabled digital health lines; KPI frameworks informed go-forward decisions on 2 of 3

Validated commercial strategy with primary research and brand health studies in HIPAA-regulated context

⊕VERIZON MEDIA

Moonshot Strategy · Jan–Dec 2021

Quantitative opportunity assessments for 5G innovation products; data-informed go/no-go investment recommendations.

5G INNOVATION VENTURES ASSESSED

Quantitative opportunity assessments for five 5G innovation products; market sizing and competitive analysis

Analyzed market trends to recommend go/no-go decisions on venture investments

◬GOPRO

Consumer Tech · Aug 2020–Dec 2021

Analytics framework for international EMEA/APAC growth strategy; GDPR data sovereignty for consumer data activation.

EMEA/APAC

INTERNATIONAL GROWTH STRATEGY

Designed analytics framework for EMEA/APAC international growth initiatives

Navigated GDPR data sovereignty requirements for international consumer data collection, activation, and attribution

◉SQUAT NEW YORK

Behavioral Health · 2019–20

Delivered seven-figure embedded product engagement for $350M ARR behavioral health company post-PE acquisition; sole tech partner from discovery through QA and handoff.

7-FIG

ENGAGEMENT VALUE · $350M ARR CLIENT

Owned product roadmap, backlog, and sprint planning; headless CMS retooling and full CX overhaul

Reduced information overload via data-driven content architecture; accelerated time-to-interaction as primary conversion KPI

Partnered across engineering, marketing, design, clinical, and operations to align stakeholders and ship on schedule

SKILLS

TYPESCRIPT

REACT

NEXT.JS (APP ROUTER)

PYTHON (FASTAPI)

POSTGRESQL

SUPABASE

PRISMA

VERCEL

API-DRIVEN ARCHITECTURE

ESLINT

GITHUB ACTIONS CI/CD

CLAUDE CODE

CLAUDE

MCP SERVER

AGENTIC AI PIPELINES

GLEAN

GROWTH MARKETING

LIFECYCLE / CRM

GTM STRATEGY

SQL

PYTHON

A/B EXPERIMENTATION

FORECASTING

DATA MODELING

CDP ARCHITECTURE

SEGMENT / MPARTICLE

EVENT TAXONOMY

ATTRIBUTION MODELING

SNOWFLAKE

DATA GOVERNANCE

OPEN SOURCE

A family of shipped developer tools, published to npm, Homebrew, and a Claude Code plugin marketplace, each verified against its own conformance suite.

rosetta-shell

Lossless shell / env translation (bash 3.2 ↔ zsh ↔ POSIX) with a CI portability gate. · npm · Homebrew

germinate

Serializable, provenance-carrying handoff protocol for multi-agent development. · npm · Homebrew

arcs skills

Claude Code plugin marketplace: root-cause analysis, state-sync, token-budgeting, toolbar triage. · plugin marketplace

→ github.com/flashesofbrilliance

EDUCATION

Oberlin College · 2002–2006

BA, Honors Studio Art

Design-first foundation: aesthetic intuition applied to data visualization and developer experience analytics.

MIT xPRO · Certificate

AI: Business Strategy

AI strategy, product instrumentation, and ML system design for enterprise and product contexts.

DEEP PROFILE

CORE STRENGTHS

Builds the intelligence layer before the category for it exists, then becomes the first and most demanding analyst of what he built. Breadth is the weapon: creative, tech, marketing, data, commercial, and security don't live in separate rooms. The rarest thing in this kind of hiring is someone who can hold all of them simultaneously and make decisions that are right across every dimension at once.

ENERGY PROFILE

ACTIVATES

First instrumentation layer: no playbook, invent the taxonomy

Cross-domain synthesis: connecting what specialists keep separate

High-stakes contexts where measurement actually matters

The moment instrumentation reveals what the team did not know to ask

Building the thing, then becoming its first analyst

DEPLETES

Pure optimization once built and proven: attention drops

ML / stat work disconnected from product decisions (no "so what")

Siloed contexts where cross-domain synthesis is not valued

Work that lives permanently in presentations, never ships

Bureaucratic consensus without clear decision authority

COMPOUND ADVANTAGES

⊗REAL-TIME BEHAVIORAL SCORING × HIGH-STAKES DECISION WINDOWS: SAME ARCHITECTURE

WTA <120ms fan scoring · live drops · 10-year gap

WTA live-event fan scoring at <120ms is the same architecture as any high-stakes, time-compressed decision window: personalized signal has to land before the window closes. Built the first version in 2016.

At WTA (2016–2018): fan at a live event → behavioral signal → personalized offer under 120ms. The system had to read intent and deliver without friction because a delayed offer during a match moment was worthless. Built the real-time architecture, the behavioral taxonomy, and the commercial partnership layer ($850K+ year-one from Twitter and Facebook).

High-stakes decision windows are the same problem class: user in a time-boxed window → behavioral signal → post-event intelligence feeds the next decision. The WTA architecture is a direct ancestor. The isomorphism is structural, not superficial.

⊗STUDIO ART × CROSS-DOMAIN INTELLIGENCE

Oberlin 2006 × breadth career

Aesthetic intelligence applied to system architecture: intelligence products that get used because they're designed to be understood, not just accurate. Creative judgment that holds across product categories and industries.

Most intelligence layers are built by people who think in one domain. The Oberlin Studio Art background is a breadth signal: trained to hold composition, material, narrative, and intent simultaneously; the same skill that makes a cross-domain intelligence layer actually work instead of producing reports no one reads.

Applied AI and forward-deployed engineering require the same skill: translating between what the system can do, what the customer needs, and what the data is actually showing. That translation is a creative act as much as a technical one.

⊗AI-NATIVE IN PRODUCTION, NOT IN DEMOS

Claude + MCP Server · ARCS

ARCS is built on Claude with a live MCP server and agentic orchestration. The onboarding question isn't "what can AI do?" but "here's what the pattern in your data is already telling us."

Applied AI roles require fluency with the actual production stack, not conceptual familiarity. ARCS runs Claude (API, tool use, extended thinking), a custom MCP server, and agentic pipelines for pre/post-decision capture. That's the stack most Applied AI / FDE candidates talk about and few have shipped.

Deployed across a full decision lifecycle: intake → analysis → structured receipt → compounding feedback. The architecture is the proof of concept.

PRODUCTIVE TENSIONS

BUILDER ↔ ANALYST

Must build to analyze well. Must analyze to build better. Breaks most job descriptions. Generates practitioner insight no pure analyst has.

AESTHETIC ↔ RIGOR

"This metric is ugly therefore unused" fires before statistical validation confirms it. Aesthetic judgment accelerates the analytical loop.

PIONEER ↔ PRODUCTION

Drawn to new territory but has the regulated-environment discipline to ship in high-stakes contexts. Valuable where new products need to be production-grade from day one.

ROLE FIT

Forward Deployed · Applied AI Engineering

Sits with customers or internal stakeholders, understands their actual workflow, and ships something useful in their context, without needing to be hand-held through the problem.

IN PRACTICE ▲ARCS was built forward-deployed by definition: no customer, no spec, no inherited infrastructure. Every architectural decision was made on contact with the real problem.

Decision Intelligence · AI Strategist

Not just measuring outcomes: designing the pre-decision capture, evidence taxonomy, and feedback loop that closes each decision against its outcome. The intelligence compounds or it doesn't.

IN PRACTICE ▲ARCS: 1,580+ PRs merged, a human approving every one. Grayscale: GTM across 10 regulated products. WTA: <120ms real-time behavioral scoring. All three required inventing the taxonomy from scratch.

Commercial layer alongside analytical depth

Growth intelligence isn't just measurement. It's knowing what the measurement is for, and being able to run the commercial decision, not just report on it.

IN PRACTICE ▲WTA: $850K year-one from first commercial partnerships. Grayscale: release-cycle time cut 35 percent with zero SEC or FINRA violations. CTF: 20% net-new revenue from data personalization. The intelligence was always in service of a commercial outcome.

THE DIFFERENTIATOR

Most candidates for Applied AI / FDE / Decision Intelligence roles have depth in one lane: analytics, or engineering, or product, or strategy. The unusual thing here is genuine breadth: creative, technical, commercial, and strategic coexisting in one person who can operate at VP level and still do the work. Add AI-native in production (not in demos), a working decision intelligence platform built from scratch, and a consistent pattern of building the intelligence layer before it had a name, and the first questions aren't about onboarding. They're about what's already in the data.

Zach Harris

Zach Harris · resume.arcs.care

2026

Agent-readable profile: zach.md