This document describes the work carried out during an internship lasting approximately three hundred hours at Sia Informatica s.r.l. The project stems from two specific needs related to Sia.Core, the platform shared by all company products and distributed across dozens of customer installations: on the one hand, the need for a safety net that allows the framework to be evolved without breaking existing functionality; on the other, the need to make the application's behavior visible in production, where a malfunction can occur far from its actual cause and with a delay that is difficult to pinpoint. The response to these needs is divided into two parallel strands. The first concerns telemetry and observability: a complete architecture based on OpenTelemetry was designed for the distributed collection of metrics, traces, and logs, with Grafana as the unified visualization and analysis platform. The second concerns automatic testing: a test infrastructure structured around unit tests, integration tests, and end-to-end tests has been introduced, aimed at ensuring the framework's correctness as the code changes. A cross-cutting element across both strands is the structured use of artificial intelligence as a pair engineer: specialized AI agents for each module, guided by procedural skills and directly connected to development tools via the Model Context Protocol (MCP), have made the propagation of telemetry and tests to new modules a replicable and consistent process, reducing hallucinations and increasing the quality of the generated code.
Il presente documento descrive il lavoro svolto durante il periodo di stage, della durata di circa trecento ore, presso l’azienda Sia Informatica s.r.l. Il progetto nasce da due esigenze concrete legate a Sia.Core, la piattaforma condivisa da tutti i prodotti aziendali, distribuita su decine di installazioni cliente: da un lato, la necessità di una rete di sicurezza che permetta di evolvere il framework senza rompere ciò che già funziona; dall’altro, la necessità di rendere visibile il comportamento dell’applicazione in produzione, dove un malfunzionamento può manifestarsi lontano dalla sua causa reale e con un ritardo difficile da imputare. La risposta a queste esigenze si articola in due filoni paralleli. Il primo riguarda la telemetria e l’osservabilità: è stata progettata un’architettura completa basata su OpenTelemetry per la raccolta distribuita di metriche, trace e log, con Grafana come piattaforma unificata di visualizzazione e analisi. Il secondo riguarda il testing automatico: è stata introdotta un’infrastruttura di test articolata su unit test, integration test e test end-to-end, volta a garantire la correttezza del framework al variare del codice. Un elemento trasversale a entrambi i filoni è l’utilizzo strutturato dell’intelligenza artificiale come pair engineer: agenti AI specializzati per modulo, guidati da skill procedurali e connessi direttamente agli strumenti di sviluppo tramite Model Context Protocol (MCP), hanno reso la propagazione di telemetria e test ai nuovi moduli un processo replicabile e omogeneo, riducendo le allucinazioni e aumentando la qualità del codice generato.
Agenti AI per web app enterprise: analisi della telemetria e automazione del testing
ZANELLA, ALESSIO
2025/2026
Abstract
This document describes the work carried out during an internship lasting approximately three hundred hours at Sia Informatica s.r.l. The project stems from two specific needs related to Sia.Core, the platform shared by all company products and distributed across dozens of customer installations: on the one hand, the need for a safety net that allows the framework to be evolved without breaking existing functionality; on the other, the need to make the application's behavior visible in production, where a malfunction can occur far from its actual cause and with a delay that is difficult to pinpoint. The response to these needs is divided into two parallel strands. The first concerns telemetry and observability: a complete architecture based on OpenTelemetry was designed for the distributed collection of metrics, traces, and logs, with Grafana as the unified visualization and analysis platform. The second concerns automatic testing: a test infrastructure structured around unit tests, integration tests, and end-to-end tests has been introduced, aimed at ensuring the framework's correctness as the code changes. A cross-cutting element across both strands is the structured use of artificial intelligence as a pair engineer: specialized AI agents for each module, guided by procedural skills and directly connected to development tools via the Model Context Protocol (MCP), have made the propagation of telemetry and tests to new modules a replicable and consistent process, reducing hallucinations and increasing the quality of the generated code.| File | Dimensione | Formato | |
|---|---|---|---|
|
thesis.pdf
accesso aperto
Dimensione
2.48 MB
Formato
Adobe PDF
|
2.48 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/111069