Software and AI

From requirement to software in production.

Platforms, applications and AI agents built with a six-stage process: each stage has verifiable outputs and a part that belongs to the client.

Reference architecture

AI is a layer of the architecture.

Agents and models are part of a platform with controlled access, data, queues and releases: not an isolated component. The diagram shows the reference structure, adapted to each project.

01ChannelsWeb appMobile appPhoneand voicePartner APIs02Access andsecurityAuthenticationRoles andpermissionsAPI gatewayLimitsand quotas03ServicesApplicationlogicAI agentsAutomationsNotifications04Data and AIDatabaseSearchand vectorsQueuesand eventsAI models05InfrastructureEnvironmentsBackupsLogs andmetricsReleasesLayered architecture: channels, access and security, services with AI agents, data and models, infrastructure.

Scroll the diagram sideways

Reference diagram. Technical choices are justified in each project’s architecture.

The software process

Six stages, with outputs and responsibilities.

1Discovery andrequirements2Architectureand UX/UI3Iterativedevelopment4Integrationsand AI5Testing, securityand compliance6Release andmonitoringiterationsSix stages of the software process, with an iterative loop between development, integrations and testing.

Scroll the diagram sideways

  1. 01

    Discovery and requirements

    Goals, users, processes and available data are put in writing, with priorities, privacy and compliance constraints.

    Outputs
    • Requirements document
    • Priority list and initial scope
    • Effort estimate and risks
    What the client provides
    • Goals and users involved
    • Access to existing processes and data
    • One contact who decides
  2. 02

    Architecture and UX/UI

    Platform structure, data model, APIs and user journeys, with interface prototypes to try before building.

    Outputs
    • Architecture diagram
    • Data model and API specification
    • Interface prototype
    What the client provides
    • Feedback on prototypes
    • Brand guidelines, if any
    • List of systems to connect
  3. 03

    Iterative development

    Built in short increments: working versions delivered at agreed intervals and reviewed at each step.

    Outputs
    • Working versions in a test environment
    • Change log
    • Up-to-date technical documentation
    What the client provides
    • Review at each increment
    • Content and test data
    • Decisions on priorities
  4. 04

    Integrations and AI

    Connection to existing systems and AI features: model choice, prompts, retrieval of company data, evaluation on test cases and guardrails.

    Outputs
    • Connectors and integration flows
    • Test cases and evaluation results
    • Guardrails and handover to an operator
    What the client provides
    • Real cases for testing
    • Documents and knowledge bases
    • System access, handed over securely
  5. 05

    Testing, security and compliance

    Functional and load tests, review of access and secrets management, assessment of the GDPR and AI Act obligations that apply to the case.

    Outputs
    • Test report
    • Security measures and data handling
    • Compliance notes to validate with a legal contact
    What the client provides
    • Privacy and compliance contact
    • Notices and appointments, if needed
    • Sign-off of acceptance tests
  6. 06

    Release, monitoring and improvement

    Go-live with separate environments, monitoring of errors and performance and an improvement cycle based on real use.

    Outputs
    • Release and rollback plan
    • Monitoring and alerts
    • List of priority improvements
    What the client provides
    • Pilot users
    • A contact for reports
    • Go-live decision

Engagement models

Three ways of working together.

01

Fixed-scope project

Scope, deliverables and calendar defined in advance.

Suited to
Clear, bounded goals
How it works
Offer with scope and deliverables; changes are handled as variations.
What is needed
Sufficiently defined requirements
02

Dedicated team

An ongoing working group that follows the client’s roadmap.

Suited to
Continuously evolving platforms
How it works
Priorities reviewed at regular intervals, with agreed capacity.
What is needed
An internal product owner
03

Retainer

A periodic commitment for maintenance, improvements and support.

Suited to
Software already in production
How it works
Hours or periodic results defined in the agreement.
What is needed
List of systems and related access

The fee is stated in the offer, after the brief is analysed.

Technology areas

Six areas, described generically.

01

Cloud and infrastructure

  • Containers
  • Separate environments
  • Networks and certificates
  • Backup and recovery
02

Backend and APIs

  • Services and queues
  • Documented APIs
  • Authentication and roles
  • Scheduled jobs
03

Frontend

  • Web apps and PWA
  • Accessibility
  • Design system
  • Responsive interfaces
04

Data and AI

  • Relational databases
  • Search and vectors
  • Language and vision models
  • Results evaluation
05

Telephony and real time

  • Voice and telephony
  • Real-time video
  • Transcription and translation
  • Live messaging
06

Security and observability

  • Secrets management
  • Logs and metrics
  • Alerts
  • Access control

Specific tools are chosen and justified at architecture stage, based on the client’s requirements, constraints and skills.

Platforms and projects

Platforms developed.

Projects and collaborations
Platform developed

LivEvent

Platform for live events and webinars in real time.

In development / pilot phase

PresenterLive roomViewersChat and questionsRecordingSimplified diagram of a live-events platform: presenter, live room, viewers, chat and recording.

Scroll the diagram sideways

Platform developed

Voice Router

Platform for voice agents and AI telephony.

In development / pilot phase

PhoneRoutingAI voice agentTools and CRMTranscriptSimplified diagram of a voice-agent platform: phone, routing, AI voice agent, tools and transcript.

Scroll the diagram sideways

Simplified, generic diagrams: they do not describe the implementation. DA COMPLETARE: description, status and role of each platform, to be confirmed with the owner

Typical use cases

Where software makes the difference.

FAQ

Software questions.

Is a large project needed to start?
No. It starts from a narrow scope, built as a prototype and assessed with real data, and is extended only if the results justify it.
How is the quality of AI answers controlled?
With test cases and evaluation before release, guardrails on allowed behaviour and handover to an operator. A total absence of errors cannot be guaranteed: critical points therefore include human approval.
How is company data handled?
The use of data with model and infrastructure providers is defined in the project and in the privacy documents before work starts. Access and secrets are managed with dedicated roles and credentials.
Who owns the code?
Ownership of code, documentation and data is defined in writing in the engagement, before work starts.
Can software accompany a physical product?
Yes: firmware, companion app and backend can be developed together with the hardware, within the product path.

Contact

A software project to assess?