Independent AI consulting for small businesses. First a close look at how the work actually runs, then a system built around it. One person designs and builds it, and your business owns it.
See the work
Four kinds of work, one rule across all of them: you keep every account and every deliverable.
An AI layer that prepares, reconciles and reviews around the software you already use. Deterministic checks first, AI for judgment, and a person signs off before anything goes out.
The system prepares, checks and reviews. A person signs off.
Every document, spreadsheet, statement and piece of paperwork across the company, copied from the drives it lives on today into one AI drive on a server you own. The originals stay where they are, untouched.
The AI drive is read-only on existing content. It can organize, rename and create new files and copies. It never rewrites an original.
Product listings, storefronts, product media, email flows and the weekly reporting behind them, corrected and kept inside your own accounts.
Listing files checked cell by cell before upload, and the weekly numbers pulled without a manual export.
Working prototypes, dashboards and internal tools, deployed on servers you control, with team accounts from day one.
A working version on a server you control, with team accounts from day one.
Once every company file sits in one AI drive, a model reads all of it and holds one working context for the business: what it sells, what it owes, what it has agreed to, what is late. It becomes the one thing that has read everything.
Three kinds of application, all built on the same AI drive and the same rule: a person signs off.
The recurring daily load, prepared and queued by the system every morning: the follow-ups, the reconciliations, the reports that come back every day. Each cycle closes faster than the last, and the people who used to chase it review it instead. Client calls transcribed and summarized before the next one.
Claude
Gmail
Google Calendar
Google Drive
Microsoft 365
Excel
Slack
Notion
Speech to text
Custom MCP connectors
Optimization loops on the marketing platforms already in place. A person sets the direction, the model runs the inference, and the workflow corrects itself as results come in, from email flows to creative to the store behind them.
Klaviyo
Figma
Shopify
Amazon Seller Central
Google Analytics
Meta Ads
Gamma
Large-scale data analysis, go-to-market models and software built to order, with the engineering behind it: backend, infrastructure, security, database and data management. Big data pipelines, data science and code-heavy systems, on servers you control.
GitHub
PostgreSQL
Data management
Excel
Python
Alpha Vantage
Custom APIs
Every application above runs on instances commissioned for the engagement, in your name, on Vultr behind Cloudflare. Set up and operated over PowerShell and Remote Desktop, with room for end-to-end models on the same machines: neural nets, multi-agent setups and machine learning, trained and served where the data already sits.
Vultr instances
Cloudflare
Docker
PowerShell and Remote Desktop
SSH and Cloudflare tunnels
GPU instances
PyTorch
Multi-agent runtimes
Four stages, in order, on every engagement. Nothing is designed before it has been observed, and nothing is relied on before it has been proven.
Weeks inside the actual work before any design. The system gets built around tasks as they are performed, not from an org chart.
Deterministic checks run before any AI judgment. Nothing reaches a filing, a live system or a customer without a person signing off.
On infrastructure you own. Accounts stay in your name; I am invited by role and never hold a password.
The first phase runs against work already completed, so every answer is known before the system produces it. Reliance is earned by measurement.
Every system starts at the size the business is today. It is built around the decisions the work already makes: what comes next, what gets checked, who signs off. At the next size those decisions are the same shape. There are more of them, and being wrong costs more. The working context, the checks and the sign-off stay. The same scripts, the same server and the same checks carry it without a rebuild.
Everything built here is meant to run without me. Once it is in your company's hands it does not need a consultant on retainer to keep going, and it does not depend on any one employee either. The goal is a self-sustaining system your company owns, understands and can adapt as it changes.
Etinger Initiatives is the practice of Federico Etinger, an independent AI consultant based in Rye, New York. Since June 2026 he has been the sole developer and architect on every engagement, across professional services, investment and e-commerce businesses.
Before this he worked in macro analysis at Morgan Stanley, at a fintech startup in Madrid and in market operations at the National Bank of Romania. He studies economics at Trinity College and is bilingual in English and Spanish.
How I work
For a project, a question or a referral, one email is enough. Replies come from Federico directly.