What does “vibe code” mean to you?

It means building with AI as a multiplier while staying engineers: every architecture, security and quality decision is verified by a person. It is not throwaway prototyping, it is software meant to run in production.

How do you handle security?

We design it in from the start on recognised frameworks (OWASP, MITRE ATT&CK and ATLAS, NIST AI RMF, the EU AI Act) and we actually apply them: from access control to defending against prompt injection in LLM applications.

What do you mean by autonomous agents with human-in-the-loop?

Agents do the work, but first they operate under human supervision: we observe their behaviour, correct it and improve it. Autonomy grows only once it is verified, and sensitive actions always keep a confirmation step.

Which AI models do you use?

We pick the right model for each task across the frontier labs (Anthropic, OpenAI, Google, Amazon, Meta, Mistral), with multi-provider routing, output evaluation and fallback. No lock-in to a single vendor.

Can the site or application be multilingual and SEO-friendly?

Yes. We work with search-engine-friendly rendering, distinct per-language URLs, hreflang and structured data. This very site is bilingual and optimised for both classic and AI search.

Tell us about the project or process you need built. We show you the approach, the architecture and a real estimate. No inflated promises.

AI-native software engineering studio

BESPOKE
SOFTWARE

We design and build complete web applications and the whole environment they live in, with security engineered in from day one and AI agents under human supervision.

Explore services Talk to us

01: Secure by design
02: Multi-model, no lock-in
03: Plugin published on WordPress.org

How we build

01Discovery
02Architecture
03App and environment
04Secure by designOWASP · MITRE ATLAS
05Human in the loop
06Deploy and observability

01Stack

Anthropic
OpenAI
ElevenLabs
n8n
Make
Qdrant
WooCommerce
Shopify
WordPress
Canva
Google Drive
19 more

02What we build

The project and its environment

We don’t just hand over an application: we hand over the system and the operating environment that keeps it running.

Bespoke SaaS, operational tools and portals, from the database to the interface.

03How we work

Autonomous agents, under human control

We build with AI agents that earn autonomy only after working under supervision: we observe, correct, improve.

01
Human-in-the-loop
Every agent action is observed and verified by a person.
02
Verify and measure
We measure behaviour and fix the edge cases.
03
Growing autonomy
Autonomy increases only where it proves reliable.
04
Confirmation on risk
Sensitive actions always keep a human confirmation.

04Secure by design

Real standards, really applied

Security is not a slogan: we design on recognised frameworks, from the web to LLM-based applications.

OWASP Top 10

Web applications

The most critical web application risks, from injection to broken access control.

OWASP API Security Top 10

APIs

API-specific risks: object-level authorization, data exposure, abuse.

OWASP ASVS

Verification

An application security verification standard to check software in depth.

OWASP Top 10 for LLM Applications

LLM apps

Risks of LLM-based applications: prompt injection, data leakage, insecure output.

MITRE ATT&CK

Threats

The knowledge base of real-world adversary tactics and techniques.

MITRE ATLAS

AI system threats

The threat landscape specific to artificial-intelligence systems.

NIST AI RMF

AI risk management

The framework for identifying and managing the risks of AI systems.

ISO/IEC 42001

AI governance

The standard for an artificial-intelligence management system.

EU AI Act

Regulation

The European regulatory framework for AI: transparency and obligations for those who build with AI.

See how we put it into practice

05AI landscape

Hundreds of models. We pick the right one.

We know the frontier labs and use each for what it does best, without tying ourselves to a single vendor.

Our operating principle

The right model for the right task: reasoning, speed or cost change the choice.
Multi-provider routing, with no lock-in to a single vendor.
Evaluation and fallback: we measure output and always keep a plan B.

Explore the stack

AnthropicClaude
The Claude family (Opus, Sonnet, Haiku).
OpenAIGPT
The GPT family.
GoogleGemini
The Gemini family.
AmazonNova
Nova models, served on Amazon Bedrock.
MetaLlama
The Llama family, open-weights.
Mistral AIMistral
Mistral models, including open-weights.

06What we’ve built

Real software, in production

Our own products and commissioned work. No invented numbers: only things that exist and can be verified.

Leomat product · E-commerce

Katapic

A SaaS that scans a WooCommerce catalogue, scores every product for visibility on Google and AI search, and rewrites the weakest listings in bulk.

Official plugin published on WordPress.org.

Visit

Leomat product · Web

leomat.it

This very site: an SPA with server-side rendering for crawlers, a blog powered by an AI editorial pipeline and AI-content disclosure as required by the EU AI Act.

Visit

Leomat product · Financial data

Portfolio intelligence for ETFs

A multi-user platform with per-user data isolation at the database level, signals put through statistical validation and the ability to abstain when the data is not enough.

Commissioned · Travel

Tour-operator management system

Tour and fleet planning, operational tasks, quotes and an AI consultant that acts on the data with human confirmation on sensitive actions.

Leomat product · Operations

Hours reporting

An hours-tracking tool integrated with project management, with monthly budgets, carry-overs and a security surface hardened after a full audit.

Commissioned · Media

Competitive SEO for publishing

A tool for analysis and a competitive content loop grounded in a real crawl of the site: proposals come from data, never invented.

07Frequently asked questions

Bespoke web software and the entire environment it lives in: the application, the infrastructure, the deployment, the security and the monitoring. From operational platforms to SaaS, up to AI agents that act on your data.

For agencies

Are you an agency? Here’s the AI department you don’t have.

White-label AI engineering arm: you keep the client and the brand, we bring the execution. Bespoke software, AI agents, integrations, operational tools. A dedicated page, made just for agencies.

Learn more

08Next step

Let’s talk about your project

Tell us what you want to build. We’ll come back with an approach, an architecture and a real estimate.

Talk to us Explore services

BESPOKESOFTWARE