Andrea Bonacci

AI Engineer LLM & Agentic Systems

I ship AI agents and RAG systems to production for companies: governed, measured, actually used.

Email me Download PDF
Cagliari, Italy
+39 393 006 3841 · andreabonacci95@protonmail.com
linkedin.com/in/andrea-bonacci · github.com/AndreaBonn · Cagliari, Italy
4+ yearsof data and AI consulting for enterprise clients
40+public repositories, 2 GitHub Marketplace Actions

Profile

From Data Scientist to AI Engineer, 4+ years of consulting for enterprise clients. Today I ship AI assistants to production that read documents, use company systems and ask a person for confirmation when it matters.

  • I build governed AI agents, RAG evaluated with metrics, real-time voice AI.
  • I come from data: pipelines, BI dashboards and NLP models, from ETL to deployment with FastAPI and Docker.
  • I teach and hire: tutor for interns, technical interviews for Data Scientists and AI Engineers, an 8-hour university course.

From Data Scientist to AI Engineer, with over 4 years of experience in data-driven consulting for enterprise clients. I design and ship production AI systems: agents and agentic workflows with MCP and tool use, LLM integration (OpenAI, Anthropic), RAG and real-time voice AI. Background in end-to-end data analysis: pipelines, BI dashboards and NLP models. Academic tutor for interns and technical interviewer for post-acquisition hiring. I also delivered an intensive 8-hour university course on AI applied to coding.

University speaker8-hour course on AI for coding, University of Cagliari, April 2026
RCE vulnerability reportedCredited in the WinBoat v0.9.1 security release
2 GitHub Marketplace ActionsAI PR Reviewer and AI Changelog Generator

Professional Experience

  1. Data Scientist internNov 2021
  2. Data ScientistApr 2022
  3. Data Analyst & DSJul 2022
  4. AI EngineerOct 2025, today
Linkalab S.r.l. · Cagliari — part of Lynx Group since Sep 2025
Nov 2021 → present
Consulting company — Internal growth
AI Engineer & Data Analyst
Oct 2025 → present
  • Production AI systems on client projects: real-time voice AI, agents for documents and accounting, RAG for regulatory compliance (details in Client Projects).
  • Tutor for university interns: training path and final evaluation.
  • Technical interviews to hire Data Scientists and AI Engineers after the acquisition.
AI Engineering focus: AI agents and agentic workflows with MCP, tool use, sub-agents and LLM integration (OpenAI, Anthropic). Production AI systems on client projects, including real-time conversational voice AI, agents for document and accounting automation, and RAG for compliance validation against regulatory corpora. Claude Code, Windsurf and Gemini CLI in production. Academic tutor for university interns. Technical interviews for hiring Data Scientists and AI Engineers post-acquisition.
AI Agents RAG Voice AI Technical Hiring
Data Analyst & Data Scientist
Jul 2022 → Oct 2025
End-to-end data analytics pipelines for enterprise clients. Interactive dashboards with Power BI, QuickSight, Tableau and Looker Studio. NLP, sentiment analysis, time series and SHAP/LIME. Deployment with FastAPI and Docker on AWS infrastructure (S3, Athena).
End-to-end data pipelines and BI dashboards (Power BI, QuickSight, Tableau, Looker Studio) for enterprise clients. NLP, sentiment analysis, time series. Deployment with FastAPI and Docker on AWS.
Python · SQL Power BI NLP AWS

Nov 2021 → Jun 2022 · Data Scientist, master's thesis internship (Italian Emotion Analysis for online reputation monitoring) followed by a contract.

Speaker — University Course
April 2026
University of Cagliari
Design and delivery of two intensive sessions (8 hours total) on "Artificial Intelligence Applied to Coding". LLM architecture, advanced prompt engineering, agentic coding with CLI tools, Model Context Protocol (MCP) integration and custom Skills creation for AI agents. Hands-on workshops where students built their first AI-assisted development workflows.
Two intensive sessions (8 hours total) on "Artificial Intelligence Applied to Coding". LLM architecture, advanced prompt engineering, agentic coding with CLI tools, MCP integration and custom Skills for AI agents. Hands-on workshops with students.
AI for Coding Prompt Engineering MCP University Lecture

Client Projects

On a contract basis for enterprise clients · details anonymized for confidentiality
AI Governance Platform for Agents
2026 · in-house product
Gateway and control plane that make AI agents usable in regulated industries: every call to models and tools goes through identity, policy decision and audit.
19 MCP tools100 policy rulesin a client trial
Technical details
  • LLM gateway with agent identity (Keycloak), policy-as-code (OPA/Rego), append-only audit trail and EU data residency.
  • Multi-tenant control plane designed and built single-handedly: human-in-the-loop approvals, policy blocks, per-agent costs (FinOps).
  • Sandboxed agent with 19 MCP tools and 100 policy rules, fail-closed multi-user isolation; enterprise chat assistant in a client trial.
AI Governance MCP OPA/Rego FastAPI · React
Conversational RAG Agent for a Manufacturer's Website
Multilingual assistant for the public website of an international industrial group, with answers that cite the source document section and a log of unanswered questions for the editorial team.
relevant context 2% → 92%faithfulness 1.004,000+ tests, 100% coverage
Technical details
  • Raised relevant context from 2% to 92% with hybrid retrieval (lexical + vector) over 800+ technical documents.
  • Measured every answer with DeepEval: faithfulness 1.00, contextual recall 0.90, no undue refusals.
  • AWS Bedrock in an EU region, pgvector, deployed on EC2; 4,000+ tests at 100% coverage.
RAG LLM Eval AWS Bedrock pgvector
Real-Time Conversational Voice AI Systems
Low-latency speech-to-speech AI conversation, with automated evaluation of interactions.
multi-provider fallbackhallucination checks
Technical details
  • OpenAI Realtime API, Gemini Live and streaming transcription (AssemblyAI), with multi-provider fallback and circuit breakers.
  • RAG knowledge base queried via function calling; hallucination checks during evaluation.
Voice AI OpenAI Realtime Function Calling
RAG & Document Intelligence for Compliance
Compliance validation of contracts, policies and playbooks against a regulatory corpus, with source citations and human review.
every finding cites its sourcehuman review with audit trail
Technical details
  • Hybrid search (BM25 + embeddings) and multi-stage agentic orchestration for structured verdicts with regulatory references.
  • Every finding links back to its source, to block hallucinations; human-in-the-loop with checkpoints and audit trail (async FastAPI, WebSocket).
RAG LLM Orchestration Structured Output Human-in-the-Loop
AI Agents for Document & Finance Automation
Automation of document and accounting processes, such as reconciliation of related documents.
output validated with PydanticERP integration
Technical details
  • Agents with function calling and validated output (Pydantic); extraction from XML, PDF and digitally signed files.
  • Deterministic matching with tolerances and confidence scoring, ERP integration and human review of uncertain cases.
AI Agent Document AI Information Extraction

Differentiators

I speak business
Business degree and a Data Science master's: I start from the company's problem, not from the model.
Proof: RAG for an industrial group
AI you can control
Agents with identity, policy and audit, built for regulated industries.
Proof: AI governance platform
I teach and hire
Tutor for interns, technical hiring interviews, an 8-hour university course.
Proof: course at the University of Cagliari

Skills

Core stack
Python AI Agents MCP RAG LLM Eval FastAPI AWS Bedrock Policy-as-Code
AI & LLM
AI Agents MCP Prompt Eng. RAG OpenAI API Anthropic API Tool Use Function Calling Document AI Human-in-the-Loop Realtime Voice Sub-agents Hugging Face Ollama LLM Eval Agent Skills DeepEval Agentic Coding Config
Governance & Security
Policy-as-Code OPA/Rego Keycloak Audit Trail Least Privilege Responsible Disclosure
Data Science
Python SQL NLP Sentiment Analysis Pandas Time Series Computer Vision
Business Intelligence
Power BI QuickSight Tableau Looker Studio
MLOps & Cloud
Docker FastAPI AWS S3 Athena AWS Bedrock pgvector Flask GitHub

Languages

Italian Native
English Fluent

Personal Projects & Open Source

github.com/AndreaBonn · 40+ public repositories

Speech-to-Text

From recorded lectures to study material: faster-whisper, a local LLM and evaluated hybrid RAG (semantic retrieval 100% vs 56% BM25 on 82 questions).

Whisper · RAG EvalGitHub

Claude Phone

Claude Code driven from Telegram, with a permission gate on tools (MCP included), human approval and audit log.

Claude Code · MCPGitHub

AI PR Reviewer GitHub Marketplace

Automated pull request review with multi-provider LLM fallback.

GitHub ActionsGitHub

Clone My Voice

Offline voice cloning and dubbing on GPU, 6 isolated TTS engines, Whisper, blind A/B tests.

TTS · CUDAGitHub

Learn Italian live

PWA with 150 lessons from A1 to C2 and spaced repetition. One of 10 live PWAs with end-to-end ownership.

React · FirebaseGitHub

AI Changelog Generator GitHub Marketplace

AI-written changelogs on every release, with multi-provider LLM fallback.

GitHub ActionsGitHub

Also: Image Generator (SDXL and FLUX), video anonymization with YOLO, Claude Code Tools Explorer, a multi-account manager for Antigravity CLI, the PWAs Meteo & Spiagge, PlotMark and BetTracker (OCR with Gemini), and the Linux tools remote shell via Telegram, Command Quiver, Sysbar and Nautilus extensions.

Critical vulnerability found and reported in WinBoat

Aug 2026
critical unauthenticated RCEfixed within 4 hourscredited in v0.9.1

Unauthenticated code execution as SYSTEM in WinBoat, an open source project. Private responsible disclosure, with runtime-verified patches and a fix shipped within 4 hours of first contact. Credited in the v0.9.1 security release.

Education

Master's Degree — Data Science & Business Informatics
University of Pisa
Thesis: Italian Emotion Analysis for online reputation monitoring
2022
Bachelor's Degree — Business Administration
University of Pisa
2018

Interested in my profile?

Email me, or download the CV as a PDF.

Email me LinkedIn PDF
Andrea Bonacci · AI Engineer Email me PDF