Writing · 17 articles
Articles about Omniscope, AI, data products, company building and air quality.
These articles come from work I have done at Visokio, technical experiments I have run myself and seven years of air-quality work around Taranto.
The publication
Browse by subject.
Each article stands on its own. The four series make it easier to follow one subject from the first practical question through to the technical and operational detail.
Verifiable and private AI analytics
Experiments with local and frontier models, agents operating Omniscope, and the controls needed to inspect the analysis they leave behind.
Building complete data products
How a useful prototype acquires data contracts, an interface, scheduling and ownership, including the point where visual tools need custom code.
Citizen science and public evidence
Taranto, benzene, public alerts and a privately operated AirGradient monitor sending measurements directly to Omniscope.
Building in the AI age
AI coding, infrastructure economics and how the technology changes the companies trying to build and sell it.
Qualità dell’aria a Taranto
The detailed account of the environmental-data work carried out with PeaceLink over time.
Series 01
Verifiable and private AI analytics.
I use language models every day and I am excited by what they can already do. These articles are about putting that capability to work inside a real analytical system, while keeping the generated queries, transformations, charts and assumptions available for a person to inspect.
What Is Verifiable AI Analytics?
Verifiable AI analytics separates model planning from governed execution, preserving the queries, calculations, lineage and evidence behind every answer.
02Why an LLM Should Not Be Your Analytics Engine
LLMs should plan and explain analytics, while governed data tools execute calculations, enforce definitions and preserve reproducible evidence.
03How AI Agents Can Operate Governed Data Tools
A practical architecture for AI agents to use governed analytics tools through permissions, bounded operations, execution records and human review.
04Local LLM Analytics: Privacy, Cost and Deployment Trade-offs
A practical comparison of local, cloud and hybrid LLM analytics across privacy, model quality, GPU cost, latency, reliability and operations.
05From Natural-Language Question to Inspectable Analytical Workflow
Follow an AI analytics request from natural-language intent through schema context, governed execution, validation, evidence and a reusable workflow.
Series 02
Building complete data products.
A prototype becomes interesting when other people start depending on it. This series follows what has to be added—contracts, validation, interfaces, automation, releases and an owner—without throwing away the visible analytical logic that made the first version useful.
From CSV Prototype to Scheduled Data Application
A practical maturity path from a CSV analytics prototype to a parameterised, monitored, permissioned and scheduled internal data application.
02Why Data Preparation, Analytics and Reporting Belong Together
Keeping data preparation, analysis and reporting connected reduces semantic drift, duplicated logic and evidence lost at tool hand-offs.
03How to Turn an Analytical Workflow into an Internal Product
Turn a working analytical workflow into an internal data product by designing users, state, actions, permissions, interfaces and operations.
04Lessons from Building Omniscope Across Multiple Technology Cycles
Seven product lessons from more than two decades of building Omniscope through desktop, web, cloud, no-code and AI technology cycles.
05Where No-Code Ends and Custom Software Should Begin
A practical decision framework for choosing visual no-code workflows, embedded Python/R/JavaScript or a separately engineered software service.
Series 03
Citizen science and public evidence.
This series records how the Taranto work grew from public files into reports, alerts, civic and institutional evidence, then a privately operated sensor. It also covers the limitations, gaps and responsibilities that need to remain visible.
How to Build a Private AirGradient-to-Analytics Pipeline
Build a private AirGradient analytics pipeline using custom ESP32-C3 firmware, local Android control and direct HTTPS ingestion into Omniscope.
02What I Learned from Analysing 80,000 Hourly Benzene Measurements
Lessons from analysing more than 80,000 hourly benzene readings in Taranto: preserve peaks, define thresholds, expose gaps and version the evidence.
03Making Institutional Air-Quality Data Usable by Citizens
How ARPA Puglia monitoring files became a public, Italian-language Omniscope instrument for citizens, classrooms, comparisons and civic investigation.
04Designing Alerts That Preserve the Evidence Behind the Claim
Design data-driven air-quality alerts with station, time, units, averaging period, reference meaning, source status and links to inspectable evidence.
05Open Hardware, Private Infrastructure and Citizen-Owned Monitoring
What citizen-owned air-quality monitoring requires across open hardware, firmware, local control, private transport, stored history and responsible interpretation.
Series 04
Building in the AI age.
The technology is moving quickly, but somebody still has to pay for the models, data centres, software and companies behind it. This series follows those economics from the practical work at the keyboard through to infrastructure, pricing and company evolution.
In Italian
The PeaceLink work, in Italian.
This article brings together the work with PeaceLink on environmental data, reports, alerts and public evidence in Taranto.
Earlier writing and source material
The wider record is still available.
The articles bring together firsthand builds, release notes, documentation, experiments and public evidence. These links lead back to the larger archives and the original work.
My Visokio articles
Product releases, data applications, local models, AI agents, verification and implementation notes.
Experiments and discussion
Work in progress, short technical observations, product arguments and public-interest analysis.
The Visokio and Omniscope story
Product history, complete data workflows, data applications and verifiable AI in one guided section.
The air-quality project
The timeline, live reports, public channels, civic evidence and open implementations behind the citizen-science series.