AWS Bedrock
Enterprise AI & Agentic RAG
Semantic retrieval, cross-encoder re-ranking, and self-improving feedback loops on AWS Bedrock & SageMaker.
Milano · Production AI & MLOps
Machine Learning Engineer.
Hi, I'm Gianluca, a Machine Learning Engineer based in Milan building production AI architectures, agentic RAG systems on AWS Bedrock, and quantized inference engines.
3+
Years Prod ML
2× AWS
Certified ML
PoliMi
Computer Eng.
#01
Enterprise GenAI & Hybrid RAG
#02
High-Throughput Quantized ML
#03
Edge Computer Vision & Video AI
#04
Full-Stack Products & Agentic OS
01 · Behind the Systems
The engineering mindset. Zero hand-offs.
Full lifecycle ownership: architecture, model training, quantization, MLOps, and live cloud deployment on AWS.
On December 4, 2023, at twenty, I started building production Machine Learning systems from scratch as the first dedicated ML Engineer at Italtel, while pursuing my B.Sc. in Computer Engineering at Politecnico di Milano.
I care about systems that are observable, maintainable, and actually used in production: from multi-step agentic retrieval on AWS Bedrock & SageMaker to low-latency ONNX INT8 pipelines and Brawn.
2× Official AWS Certifications
Machine Learning Specialty (MLS-C01) & Cloud Practitioner (CLF-C02)
Gianluca Meneghetti
Milano · Driven by hard problems
Sustenpass · Street Triple 765 RS
Swiss Alps · 130 hp
2,224 M · CH
01B · Core Pillars
Four domains. One engineering standard.
From enterprise agentic retrieval and quantized ONNX inference on AWS to independent SaaS products and Linux systems.
ONNX INT8
High-Throughput Quantized ML
Ingestion, feature engineering, and quantized ONNX INT8 inference serving millions of records in production.
getbrawn.app
Brawn · Hybrid Athlete Platform
Unified strength, combat sports, endurance, and recovery platform built around a single athletic load model.
Systems & OS
Agentic OS & Linux Systems
Recursive CLAUDE.md routers across Obsidian and low-level Linux hardware integration.
02 · The Collection
Things I've actually shipped.
Separated clearly between enterprise production ML systems and independent products.
Part 01 · Enterprise Production ML (3 Systems)
01 · ENTERPRISE GENERATIVE AI
Hybrid Search & Multi-Agent
Enterprise Agentic RAG Platform
Multi-step agentic retrieval-augmented generation over enterprise knowledge bases. Hybrid semantic + lexical reranking, persistent vector indexing, and automated evaluation loops for citation accuracy and hallucination control.
02 · QUANTIZED ML INFERENCE
High-Throughput Stream
High-Throughput ML Inference Engine
Production-grade classification pipeline for high-volume operational and observability telemetry. Ingestion, automated feature pipelines, and post-training quantization for low-latency stream classification on cost-effective CPU infrastructure.
03 · EDGE COMPUTER VISION
Multi-Stream Video
Industrial Video Analytics & Safety Vision
Real-time multi-stream video intelligence for operational safety in field conditions: worker safety, PPE compliance, and critical event detection across concurrent RTSP feeds with deterministic low-latency edge inference.
Part 02 · Products, Startups & Open-Source Systems
Ref. 004 – 007
04 · FOUNDER · AI SAAS
getbrawn.app
Brawn · Hybrid Athlete Platform
Unified training, nutrition, and recovery platform for hybrid athletes who lift, fight, and run, with interval combat timers and AI coaching.
05 · LINUX HARDWARE DRIVER
libfprint · USB / PAM
EH57E · Linux Fingerprint Driver
A libfprint driver for the EgisTec sensor inside Samsung Galaxy Book power buttons. USB protocol reconstructed; enrollment and PAM login working.
06 · TELEMETRY ANALYTICS
Unsupervised ML
Motorsport Telemetry Suite
Unsupervised ML driving coach: K-Means segments lap data into 5 objective driving phases from speed, throttle, brake, and RPM.
07 · ADAPTIVE SCHEDULING & VOICE OS
Multi-Agent
PrepWise & Batcave Executive OS
Co-founded PrepWise (AI study planner with spaced repetition around real deadlines) alongside Batcave, my voice-driven personal executive assistant.
03 · Agentic OS & Specifications
Interactive System Graph & Full-Stack Specs.
Drag any node to explore how my recursive CLAUDE.md / AGENTS.md routers, Obsidian vault, AWS Bedrock pipelines, and products connect.
LIVE INTERACTIVE TOPOLOGY · CLAUDE.MD ↔ AGENTS.MD ↔ AWS BEDROCK
DRAG NODES · HOVER TO INSPECT CLUSTER
ACTIVE NODE
CLAUDE.md ↔ AGENTS.md Core Router
Paired hierarchical context routers orchestrating Claude Code, Codex, Antigravity, Obsidian Vault, and production ML workflows.
03B · Technical Architecture
Full-Stack ML Specifications.
Production stack across cloud GenAI, quantized inference, computer vision, and agentic tooling.
MLOps & Cloud
AWS BedrockAWS SageMakerS3 · ECR · ECSLambdaDockerKubernetesMLflowCI/CD
ML / DL & Vision
PyTorchTensorFlowONNX Runtime INT8TensorRTLightGBMscikit-learnKerasCV
AI Agents & Tooling
CLAUDE.md ↔ AGENTS.mdObsidian VaultLangGraphLangChainHybrid RAGFastAPIn8n
Languages & Data
PythonSQLPolarsPostgreSQLClickHouseChromaDB
04 · Provenance
Where I've been building.
Shipping enterprise ML systems full-time while completing a full Computer Engineering degree at Politecnico di Milano.
· Work
Machine Learning Engineer · Italtel
Designing and deploying production Machine Learning systems across real-time computer vision, high-throughput operational inference, and enterprise GenAI architectures from problem definition to live cloud operations.
Founder & Lead Engineer · Brawn
Designed and built the unified training, nutrition, and recovery platform for hybrid athletes who lift, fight, and run (getbrawn.app).
· Study
B.Sc. in Computer Engineering
Ingegneria Informatica at Politecnico di Milano, pursued alongside full-time Machine Learning engineering.
AI & ML Specialization · ITS Academy Angelo Rizzoli
Two-year specialization in applied Machine Learning, Deep Learning, Data Engineering, and Cloud MLOps. Graduated with 100/100.
05 · The Other Hours
The racing line & discipline.
130hp on alpine passes, mechanical horology, kickboxing, and progressive overload.
AUTODROMO DEL MUGELLO · 5.245 KM · 15 TURNS
AMBIENT · 3 RIDERS · ARRABBIATA
BIKE · TRIUMPH STREET TRIPLE 765 RS · 130 HP
NOLAN X-804 RS ULTRA CARBON
01 · APEX
Motorcycles
Street Triple 765 RS
130hp, mountain passes, early mornings. Confidence and precision aren't opposites.
02 · OVERLOAD
Kickboxing & Gym
Discipline > motivation
Show up when you don't feel like it. Progressive overload: the body responds to data if you track it properly.
03 · CALIBER
Mechanical Watches
±4s a day · No battery
Hundreds of mechanical components working in harmony. The engineering is the art.
04 · ALPS & BOOKS
Skiing & Reading
Read terrain, commit
No hesitation halfway through a turn on snow, and high-bandwidth books to download new mental models.
Apex · Onboard765 RS
