Skip to content
Frank Vitetta

Frank VitettaAI implementation consultant. London.

I help teams get their data ready for AI and run AI privately, on their own machines.

Read the field notesHire me

  • Table formats
  • Measurement
  • Sourced claims
  • No jargon

Introduction

I come to this subject from the marketing data side. I am not a data engineer and I do not pretend to be one.

As a consultant I help teams get their data ready for AI. I also help them run AI privately on their own machines when that matters. On this site I explain the infrastructure underneath in plain language. I read the specifications, documentation, standards and research. Then I write down what they say without the jargon. I cover data platforms, table formats, analytics engines, data quality and measurement. Marketers, analysts and decision makers are asked to rely on these things and rarely get them explained.

Every factual claim on this site links to its source. I say so when something is my opinion. Read more about me and how I work.

Selected writing

Field notes

Each note has numbered sections, a diagram, a comparison table, a plain-English summary and a list of the sources I used.

  1. N° 001

    Open table formats explained: Iceberg, Delta Lake and Hudi

    What a lakehouse is. How Apache Iceberg, Delta Lake and Apache Hudi turn files into tables. Where each format stands in October 2026.

    Table formats6 min read

  2. N° 002

    Prompt length vs brand mentions: 8,555 ChatGPT responses

    8,555 ChatGPT responses by prompt length. Mention rate peaks at 35 to 69 characters and is 6.5% at 120 and over. One weak bucket rests on 40 responses.

    Measurement8 min read

All field notes / Index of terms

A plate from my notebook

PlateConceptual drawingScroll sideways

Where quality checks sit in a marketing data pipeline Data flows from sources, through collection, into storage and modelling. From there it feeds two branches: dashboards that inform decisions, and a retrieval index that feeds an AI answer. A band of quality checks sits above every stage. A dashed line shows that a defect entering at the sources travels all the way to the AI answer. Quality checks at every stage complete / unique / timely / valid / accurate / consistent Sources ads, CRM, web Collect load, import Store, model tables, joins Dashboards Decisions Retrieval index AI answer A defect that enters at the source travels to every later stage
What it shows A marketing data flow with quality checks at every stage. The dashed line shows how a defect that enters at the source travels to every later stage. That includes an AI answer.

A project of mine

Code Scout: a free AI coding agent that can run on your own machine

AI coding got expensive fast. Code Scout is my answer to that. It is a native desktop app that can run local language models, so your code stays on your machine.

It is free to download and needs no sign-up. It is in alpha. LLM Scout, the company I founded, makes it. There are builds for macOS, Windows and Linux.

Download Code ScoutHow it works

Audit, implementation, training

Private AI and data readiness consulting

Your team may want to use AI without sending its code or data to someone else's cloud. I can help you get the data ready and run AI on your own machines.

I offer three services. The first is an AI data readiness audit and strategy. The second is private AI implementation with existing tools. The third is training and workshops. Every job starts with one paid working session. I quote for it in writing before you book. I do not build pipelines or offer data engineering.

See how I work