I design AI adoption and governance for large organisations and take AI agents from prototype to production: use-case portfolios and value tracking, governance that sits inside delivery, process redesign with process mining, and data governance that owners actually adopt. I work in Italian, English, Spanish and French.
Every repository here is built the same way: a deterministic core with tests in CI, an agent only where language helps, acceptance criteria written before the code, red-team and injection cases in the evals, and synthetic data so everything can be public.
- agent-deployment-engagement-kit - A customer-facing AI agent from discovery to the sponsor's widen decision in 12 weeks: bilingual reference agent, escalation policy, offline and red-team gates, pilot metrics, steering note.
- process-mining-to-ai-redesign - From an event log to an AI-first redesign proposal: process mining, AI-fit with hard autonomy caps, an explicit value case, and an agent that drafts the process owner's memo.
- data-governance-stewardship-kit - Data governance as an operating model: CDE register with owners and stewards, DAMA-aligned quality rules as code, owner scorecards, issue SLAs, adoption stages.
- governed-agents - Governed multi-agent runtime: a policy engine decides every tool call, external actions wait for a named approver (four eyes), every step is audited, and agents are evaluated on what they did, not only what they wrote.
- ai-engineering-framework - A local-first engineering framework for governed AI services: reference architecture, patterns, standards as CI checks, use-case intake and a service template.
- ai-adoption-tracker - Evidence for an AI review committee: which pilots to scale, iterate or stop, and why. MVP with its product and team pack.
- agentic-procurement-intake - From 'I need a consultant' to a request procurement can approve on first review: an intake agent prepares, people decide. MVP with its product and team pack.
- governed-llm-gateway - Internal OpenAI-compatible gateway that routes requests to the lowest-cost suitable model (Claude, Azure OpenAI, Mistral AI, or local open-weight), masks personal data, enforces team budgets, and reports chargeback without storing content.
- governed-ai-platform - Reference deployment of a governed AI stack: LLM gateway, MCP evidence server and multi-agent service as hardened containers, with secrets, Prometheus monitoring and alerts, pinned versions, a runbook and an end-to-end test in CI.
- feedback-to-roadmap - From a pile of user feedback to evidence per theme and an explained roadmap change proposal: the tool proposes, the product owner decides. MVP with its product and team pack.