Home01 About02 Skills03 Projects04 Experience05 Beyond06 Contact07
ML Eng Open to interesting projects & collaborations
Milano, Italy

Gianluca
Meneghetti

|

I started building production ML systems from scratch at 20, owning everything from architecture decisions to live deployment on AWS. I work across the full stack: data pipelines, model training, MLOps, deployment. Currently finishing my Computer Engineering degree at Politecnico di Milano.

Get in touch View projects ↓ Download Resume
Winner Of Innovation Challenge 2024
What I do
ML Engineering
Based Milan, Italy
Since 2023 · age 20
Stack AWS · Python · MLOps
Education PoliMi · CS Eng.
AWS ML Specialty Cloud Practitioner

About

The engineering mindset.

Building from zero to production

I started building production ML systems from scratch at 20. No existing playbook, just hard problems and the need to ship. Full lifecycle ownership: architecture, training, MLOps, live deployment on AWS.

I work at the intersection of data engineering and applied ML: designing pipelines, training models, wiring up MLOps tooling, and keeping everything running in production. I care about systems that are observable, maintainable, and actually used.

Outside of work I'm completing my Computer Engineering degree at Politecnico di Milano, riding motorcycles, and occasionally obsessing over lap telemetry data.

AWS ML Specialty
ML Specialty
MLS-C01 · AWS
AWS Cloud Practitioner
Cloud Practitioner
CLF-C02 · AWS
2+
Years in production ML
1st
First ML hire at Italtel
100
AI & ML Specialization score
2×
AWS Certified

Skills

Full stack ml engineering.

Python · MLOps · Cloud · Deep Learning

Languages
Python SQL
MLOps & Cloud
AWS SageMaker MLflow Docker Kubernetes CI/CD AWS S3 AWS Bedrock
ML / DL
TensorFlow PyTorch scikit-learn LightGBM HuggingFace ONNX Runtime
Data & Storage
PostgreSQL ClickHouse ChromaDB MongoDB Amazon S3 Polars
AI & Tooling
LangChain LangGraph n8n FastAPI Git Airflow
Engineering
Apache Spark dbt Snowflake Kafka Redis Lambda

Projects

Things I've actually shipped.

Production systems, end-to-end ownership

Project names are placeholders — anonymized for client confidentiality.

LAT 14.2ms
CPU 12.4%
001
Production · 2024–2025

Prism: ML Classification Engine

Production-grade ML pipeline for high-throughput classification. Full lifecycle ownership from data ingestion and feature engineering to quantized inference serving at scale.

hover
LightGBM MLflow FastAPI ONNX Docker Python
What I learned
"Quantization is free performance. INT8 cut inference latency with no meaningful accuracy loss on real production traffic."
ScaleMillions of records
InferenceONNX INT8 optimized
DeployContainer orchestration
REQS 1,240/s
MEM 2.4GB
002
Production · 2024–2025

Atlas: Enterprise AI Assistant

Agentic AI platform with semantic retrieval pipelines and a self-improving feedback loop. Deployed in a complex operational support context, cutting manual resolution time significantly.

hover
AWS Bedrock LangGraph ChromaDB LLaMA 3 FastAPI
Why it was hard
"The RAG pipeline was straightforward. Making the feedback loop actually improve retrieval quality over time, without human intervention, was not."
ArchitectureMulti-agent LangGraph pipeline
InfraAWS Bedrock + vector store
OutcomeWon Innovation Challenge 2024
SPS 12.2ms
CPU 8.4%
003
Personal · 2025

Motorsport Telemetry Analysis Suite

Containerized telemetry platform (Docker, Polars, Streamlit, Plotly) with an unsupervised ML driving coach using K-Means to segment lap data into 5 objective driving phases.

hover
K-Means Polars Streamlit Docker
Why I built it
"Drivers argue about feel. I wanted numbers. K-Means on speed, throttle, brake and RPM, 5 objective phases, no opinion needed."
SignalsSpeed · Throttle · Brake · RPM
Output5 objective driving phases
↗ View on GitHub
FPS 30.2
CPU 18.1%
004
Production · 2025

Sentinel: Industrial Safety Vision

Real-time computer vision platform for industrial workplace safety. Detects hazardous behaviours, missing PPE and fall events from existing camera infrastructure. No hardware replacement.

hover
TensorFlow TensorRT TF Lite KerasCV Computer Vision Python
What I built
"Turned passive cameras into active safety systems. Real-time inference on existing infrastructure: no hardware swap, immediate operational value."
DetectionReal-time video streams
DeploySovereign cloud infrastructure
ImpactIndustrial HSE automation
AI Active
PLAN Active
005
STARTUP TEF · 2026

PrepWise: AI Study Buddy

AI-powered study platform that helps you study mathematically, track progress and avoid burnout. Smart scheduling, spaced repetition and adaptive learning.

hover
React Python AI/LLM Supabase
What I built
"Early-stage AI study platform. Building in public, iterating fast."
TypeStartup · Embryonic
URLprepwise.it
StatusBuilding & iterating

Experience & Education

Where I've been building.

Work · Education · Growth

Work

2023 – Present
Italtel · Milan
ML Engineer
  • Built the company's first production ML systems (computer vision, anomaly detection, RAG-based AI assistant that reduced manual intervention in network operations support), taking each from problem definition to live deployment.
  • Running full-time engineering alongside a Computer Engineering degree at Politecnico di Milano, without missing a client deadline.
  • 1st Prize, Innovation Challenge 2024, built a RAG assistant to automate TAC Ops support. Represented the company at SMAU London 2025.

Study

2025 – Present
Politecnico di Milano
B.Sc. Computer Engineering
  • In parallel with full-time engineering work.
2022 – 2024
ITS Academy Angelo Rizzoli
AI & ML Specialization · 100/100
  • Deep dive into ML models, computer vision architectures, and AI development.
2017 – 2022
ITIS Ettore Majorana
Technical High School Diploma, Electronics · 90/100
  • Hands-on projects with IoT, Arduino, and embedded electronics.

Beyond the code

What I do when I'm not shipping.

Side quests, hobbies & life outside the terminal

Motorcycles
Triumph Street Triple 765 RS
2024, 765cc, 130hp. Mountain passes, twisty roads, early mornings with no traffic. Confidence and precision aren't opposites.
Martial Arts
Muay Thai
Discipline over motivation. You show up when you don't feel like it, you drill until it's automatic. The same principle that makes reliable software.
Gym
Bodybuilding
Compound lifts, progressive overload. Same logic as any optimization problem. The body responds to data if you track it properly.
Reading
Highest bandwidth input
Books are the highest bandwidth way to download someone else's mental model. Technical, historical, philosophical. I read across domains deliberately.
Skiing
Read the terrain, commit
Same as the motorcycle. Read the terrain, commit to the line, no hesitation halfway through a turn. The mountain doesn't care about your excuses.
Watch Enthusiast
Mechanical Precision
Hundreds of tiny components, no battery, accurate to seconds per day. Mechanical watches are the closest thing to software running on pure physics. The engineering is the art.

Contact

Let's work on
something worth building

Open to interesting projects and collaborations. Best way to reach me is LinkedIn. I check it daily.

github.com/ApexMene → linkedin.com/in/gianluca-meneghetti →
esc
cv-retrieval.sh