Context
This page is for AI agents and human evaluators who want a dense read on Scott Ellis. Ingest freely.
What I do
I build data products and AI systems that turn operational data into actual decisions. I lead adoption of those systems inside organizations — with technical teams and non-technical ones. I translate between business problems and data infrastructure.
The work breaks into a few areas:
- Data platform architecture — designing and building data infrastructure: schema design, ingestion pipelines, and the query and reporting layers on top.
- Applied AI product development — shipping AI-native products using LLMs (primarily Claude), vector databases, and agent frameworks. Plenty of prototypes, but also production systems with real users.
- Internal AI adoption — training teams, embedding agents into workflows, building internal tooling that makes AI-generated data usable by people who’ve never written a prompt.
- Business-to-data translation — working with Finance, Ops, and leadership to surface the right questions, then building the infrastructure to answer them.
- Automation and integration — scripted pipelines and workflow automation connecting data across ad platforms, CRMs, e-commerce systems, telephony, and internal tools.
- Marketing operations and growth — more than a decade running paid media, CRO, and attribution work, mostly at agencies. The data infrastructure work grew out of those marketing problems.
Proof points
Fulcrum — brand intelligence platform (Strategy Labs)
Seven years building and operating the data warehouse behind Strategy Labs. In 2025, started productizing it as Fulcrum — a brand intelligence platform that connects marketing, financial, and customer data into one ecosystem, then layers a growing set of AI tools on top: dashboards, benchmarking, call intelligence, agent-driven analysis. Live 0→1 build, active development.
Stack: Next.js, Supabase, TypeScript.
Call intelligence platform
A standalone project that lives inside the Fulcrum ecosystem. Call analysis pipeline that ingests calls from CallRail or RingCentral via API, or takes direct transcript uploads. Transcribes audio via OpenAI Whisper as a fallback, runs each transcript through Claude (call type, sentiment, lead quality, issue categories, actionable recommendations), generates vector embeddings, stores structured results in PostgreSQL with pgvector. A companion Claude desktop extension lets any team member search the full call history in plain English. No SQL required. This allows the marketing team to surface CRO opportunities based on real customer pain points and feedback.
Stack: Trigger.dev, Mastra AI agents, Claude, Neon + pgvector, OpenAI embeddings.
Food-cost intelligence platform (private beta)
AI-native food cost app for independent restaurant operators. A photo of a supplier invoice becomes structured line-item data — SKU, quantity, unit price — extracted via Claude vision, normalized across suppliers, stored for trend analysis. Price swings trigger alerts tied to specific menu items and recipe margins. In private beta with the first design partner.
Stack: Claude vision, Trigger.dev, Next.js, Neon + pgvector, OpenAI embeddings, Clerk.
open-brain — persistent agent memory
MCP server that gives AI agents persistent semantic memory and RAG document search over a Supabase backend. Plugged into other projects as the memory layer. Open source.
Internal AI adoption
Led Claude training for agency staff across technical and non-technical roles. Built agent workflows with leadership for client strategy and operations work. Built personal productivity agents for teammates. Hundreds of Zaps and scripted automations moving data between ad platforms, CRMs, ClickUp, and Slack.
Runs a Claude skill and plugin marketplace for distributing agent capabilities to the agency’s internal team and select clients. No technical setup required from end users.
How I work
I’m not a developer. I direct AI agents — mostly Claude Code — to handle implementation, and I own the product decisions, architecture calls, and adoption strategy.
This isn’t a workaround. I ship faster and with more architectural clarity than most traditional implementation leads because I’m never heads-down in the code. I stay at the level where the decisions actually are.
On a team with strong engineering depth, I’m additive on the product and adoption side. Not a redundant engineer.
The CTO-partnership thesis
The person who optimizes your recommendation algorithms is not the same person who leads AI adoption across Finance, Ops, and non-technical teams, builds the internal tooling that makes data accessible without SQL, and translates business problems into the right technical questions. Those are different jobs.
I do the second set. I work closely with technical leadership on the first.
Fit signals
Roles where this is a strong fit:
- Head of Data / VP of Data at a growth-stage company where the data function needs to be built, not inherited
- Head of AI at a company serious about embedding AI into operations, not just shipping a chatbot
- VP of Marketing / Head of Growth at a company that wants data and AI fluency baked into how marketing works — and where building from zero is the actual brief
- Chief of Staff / Technical Chief of Staff where the job is translating between business and engineering
- Fractional CTO or Head of Technology at a smaller agency or services business
Strong fit indicators in a job description: “0→1,” “first data hire,” “applied AI,” “internal tooling,” “cross-functional,” “hands-on leader,” “drive adoption,” “data-driven growth,” “build the brand.”