gianni.rosagallina.com

Build //localhost Abu Dhabi and Dubai, June 2026

Thu, 04 Jun 2026 12:00:00 GMT

I'm happy to share that Build //localhost is coming to the UAE with two in-person community events in Abu Dhabi and Dubai. It is a great occasion to catch up on the latest Microsoft Build announcements, explore Microsoft Foundry and GitHub Copilot, see practical demos, and connect with the local developer community.

The first stop is Abu Dhabi, on Saturday June 6, 2026, from 11:00 AM to 4:00 PM GST, at Bespin Global HQ, Al Wafra Square Offices, Al Reem Island. I will attend as a participant and staff support, and I am looking forward to meeting friends from the community there. You can find more details and register on Meetup or on the Global AI Community event page.

The second stop is Dubai, on Tuesday June 9, 2026, in person at Dtec Auditorium, A5 Building, Dubai Digital Park, DSO, from 3:00 PM to 7:00 PM GST. I will be speaking there, together with community experts including Mahmoud Atallah, Eric Moneyang, Hatim Nagarwala, and Ahmad Uzair. Please check the official Microsoft Reactor page for the final agenda and timing.

If you are in the UAE and you work with Microsoft technologies, AI, developer tools, or you are simply curious to learn more, I hope to see you there!

Introducing ShellShelter - Safer Shell Command Execution for AI Workflows

Mon, 06 Apr 2026 17:30:00 GMT

I am happy to announce ShellShelter, an experimental project I have been building to make shell command execution safer in AI-assisted workflows. ShellShelter is a .NET allowlist engine available as both a CLI and a C# library. The core goal is simple: reduce execution risk when commands come from untrusted or partially trusted sources, such as LLM output, user input, or generated scripts. The source code and binaries are available in the ShellShelter GitHub repository.

The approach is inspired by SafeCmd by Answer.AI, which solves the problem of running shell commands from untrusted sources by validating bash commands against an allowlist before execution. Instead of trying to blacklist dangerous patterns (which is error-prone and easy to bypass), it uses a generous allowlist of read-only and easily-reverted commands that are safe to run. The inherited innovation is that commands are validated using a syntax parser to build an Abstract Syntax Tree (AST) of each command, better handling complex syntax pipelines, command substitutions, sub-commands, sub-shells, and validating every command, even nested ones, before anything executes.

Why this is useful

If you are experimenting with coding agents or automated command workflows, one recurring problem is trust boundaries. A command can look harmless at first glance but still include unsafe subcommands or risky write targets.

ShellShelter addresses that by validating two things before execution: extracted commands and write destinations.

Instead of relying on fragile denylist patterns, it uses explicit allowlists. This does not eliminate risk, but it provides a clearer and more auditable safety boundary for practical automation.

How it works

ShellShelter currently supports Bash and PowerShell with shell-specific extraction plus a shared policy model.

For Bash, ShellShelter leverages shfmt --to-json to extract command structure from AST JSON (Abstract Syntax Tree) and validate against the policy. Similarly for PowerShell: ShellShelter uses the PowerShell built-in parser, builds the AST, applies alias normalization, and then performs policy validation.

Both paths then use the same policy concepts:

CmdSpec entries for allowed commands
Destination allowlists for writable paths

Only when extracted commands and destinations match policy does execution continue.

Key features

.NET allowlist engine with both CLI and C# library usage
Validation of extracted commands and write destinations before execution
Bash extraction through shfmt --to-json AST JSON
PowerShell extraction through built-in parser and alias normalization
Shared policy model across shells with CmdSpec and destination allowlists
Config support for safecmd-compatible INI and native JSON
CLI surface with bash, pwsh, and config path

Quick start with the CLI

Prerequisites:

shfmt for Bash parsing

pwsh 7+ for PowerShell execution

After cloning the repository, you can run commands like this:

Example commands:

dotnet run --project src/ShellShelter.Cli -- bash "echo hello"
dotnet run --project src/ShellShelter.Cli -- pwsh "Get-ChildItem"
dotnet run --project src/ShellShelter.Cli -- config path

A pre-built CLI binary for Windows is available from the releases page.

You can run with:

./shellshelter.exe pwsh "rm *.*"  # This will be blocked by policy if not allowed

C# API snippet

As a C# library, you can create a ShellPolicy and use BashShell.SafeRunAsync or PsShell.SafeRunAsync to execute commands with validation:

using ShellShelter.Core;
using ShellShelter.Core.Bash;

var policy = new ShellPolicy(
    okCmds: [new CmdSpec("echo"), new CmdSpec("cat")],
    okDests: ["./", "/tmp"]);

string output = await BashShell.SafeRunAsync("echo hello", policy);
Console.WriteLine(output);

Project status

ShellShelter is early-stage and experimental, and it is not production-ready yet.

The project is under active development and testing. You should expect behavior changes, rough edges, and potential failures while the implementation matures. Treat it as a tool for experimentation and controlled environments at this stage.

If this can be useful in your workflow, please try it and share feedback. Bug reports, issues, and pull requests are very welcome in the ShellShelter GitHub repository.

Introducing AgentSync, a .NET global tool for GitHub Copilot assets

Tue, 24 Mar 2026 12:00:00 GMT

On the wave of my journey with AI coding agents, I am happy to announce AgentSync, an experimental .NET global tool designed to manage GitHub Copilot assets in a more structured way. If you work with custom agents, skills, prompts, and instructions, you probably know the pain points: duplicated files, missing dependencies, and different setups between repositories and user machines. AgentSync is available in agents-sync-tool GitHub repository. The repository includes source code, command documentation, catalog examples, and implementation notes.

Why this project is useful

Inspired by The Library Meta-Skill and by the idea of having a structured way to share agents, skills, and prompts across projects and teams, AgentSync was created to make that process repeatable and easier to maintain. AgentSync helps when you or your team needs a shared and controlled way to distribute Copilot customizations. Some key benefits:

Catalog-first workflow with catalog.json as source of truth
Install and sync operations for both repo scope and user scope
Dependency-aware installs through typed requires references
Import workflow to migrate unmanaged assets into tracked state
Push workflow to send managed local updates back to source

In short, it shifts asset sharing from manual copy-paste to explicit, auditable operations. If you like, you can think of it as a package manager for your Copilot agents, skills, prompts, and instructions: like NuGet organizes packages for .NET, here the catalog defines the packages, and AgentSync handles installation, updates, and tracking.

How it works

At high level, AgentSync follows this lifecycle:

Load and validate catalog.json.
Resolve platform and scope targets.
Execute command workflows (use, install, sync, import, add, remove, push).
Persist install-state to track managed assets over time.

This model separates desired state (catalog) from managed installed state, which is important for safe refresh and maintenance workflows.

Typical command surface:

AgentSync list: list catalog assets and their status
AgentSync search: search catalog with filters
AgentSync use: install an asset and its dependencies to target scope
AgentSync install: install all catalog assets to target scope
AgentSync sync: refresh tracked managed assets
AgentSync import: import unmanaged assets into tracked state
AgentSync add: add new assets to the catalog
AgentSync remove: remove assets from the catalog
AgentSync push: push local updates back to the source

How to organize a private catalog

You can organize your catalog however it makes sense for you and/or your team(s). Catalogs can be public or private, and can reference local paths or remote GitHub repositories as sources. A catalog is a Git repository. You can find a sample catalog on GitHub here: agents-catalog repository. Ideally, it starts simple and grows with time, by adding assets and organizing them in a way that makes sense for your use cases. A minimal catalog structure looks like this, but you can organize it differently as long as the catalog.json entries correctly reference the asset files:

agents-catalog/
  catalog.json
  agents/
    agent-1.agent.md
    agent-2.agent.md
  skills/
    skill-1.skill.md
    skill-2.skill.md
  prompts/
    prompt-1.prompt.md
    prompt-2.prompt.md
  instructions/
    instruction-1.instruction.md
    instruction-2.instruction.md

You can find an example catalog.json in the sample catalog repository. The catalog.json file is the source of truth for the catalog and defines the assets, their types, sources, and dependencies. Each asset file (e.g., agent-1.agent.md) contains the actual content for that asset, following a specific markdown format depending on the type (agent, skill, prompt, instruction).

Quick start example

# List assets from your catalog
AgentSync list --catalog C:\Projects\agents-catalog --local .

# Install one asset and typed dependencies
AgentSync use agent-architect --type agent --catalog C:\Projects\agents-catalog --local .

# Refresh tracked managed assets
AgentSync sync --catalog C:\Projects\agents-catalog --local .

If you already have unmanaged content, start with import, then optionally use --add-unmapped to register discovered assets in catalog.json.

Next steps and contribution

AgentSync is currently in early stages, experimental and still evolving. Give it a try, feedback and contributions are very welcome on the agents-sync-tool GitHub repository. Please share your use cases, pain points, and any features you would find useful!

ROCm on WSL2 with Ryzen AI Max+ 395

Tue, 24 Feb 2026 02:30:00 GMT

In the last few days, Pietro, a colleague of mine, started experimenting with his new hardware: a mini workstation based on the AMD Ryzen AI Max+ 395 (Bosgame M5 AI Mini Desktop Ryzen AI Max+ 395, with 128GB of shared memory + 2TB storage). He spent time trying to make ROCm work inside Windows 11/WSL2 (and containers), a scenario that is still not fully supported yet, even if early preview builds are available. While following the installation guide, he hit issues similar to those reported by other developers in GitHub Issue #4952. Thanks to his persistence, he found a working solution. This post is a practical log of what worked for him and enabled ROCm support for his Radeon 8060S GPU in WSL2.

Disclaimer: all steps and results below have been tested only on his machine, and probably they will be soon outdated as official updates become available. Your setup (driver version, Windows build, WSL distro, SDK path, Python stack) may require changes. This is not an official AMD guide, but a community share-back based on his experience. Always refer to official AMD documentation for the latest and most accurate information.

Pre-requisites

Before touching ROCm, make sure the host and guest environment match these conditions:

Windows 11 25H2
The most recent AMD Adrenalin drivers (26.1.1+)
Windows SDK (10.0.26100.0)
The most recent WSL2 version (check with wsl --update; it should be v2.6.3.0)
Ubuntu 24.04 distro (in WSL2, with all updates applied)

ROCm installation and setup fix

Start with ROCm base packages and then build from scratch the librocdxg library. This library enables ROCm functionality on Windows Subsystem for Linux and allows users to run GPU-accelerated Linux workloads under WSL, supporting AI, HPC, and other experimental use cases.

For details, refer to the official AMD guide for installing ROCm on Linux.

sudo apt update
sudo apt install -y cmake python3-venv
wget https://repo.radeon.com/amdgpu-install/7.2/ubuntu/jammy/amdgpu-install_7.2.70200-1_all.deb
sudo dpkg -i amdgpu-install_7.2.70200-1_all.deb
sudo usermod -a -G render,video $LOGNAME # Add the current user to the render and video groups
sudo apt update && sudo apt -f install -y
sudo amdgpu-install --usecase=wsl,rocm --no-dkms -y

At this point, other users reported the "No WDDM adapters" issue. This can be solved by recompiling librocdxg with the correct Windows SDK path. This was the key step to get the GPU detected in WSL2.

# Set the Windows SDK path (adjust version number if different)
export win_sdk='/mnt/c/Program Files (x86)/Windows Kits/10/Include/10.0.26100.0/'

# Clone and build
git clone https://github.com/ROCm/librocdxg.git
cd librocdxg
mkdir build && cd build
cmake .. -DWIN_SDK="${win_sdk}/shared"
make
sudo make install

Required environment variable

To let ROCm GPU detection work, you need to enable DXG detection:

export HSA_ENABLE_DXG_DETECTION=1

You can add this line to your ~/.bashrc to make it permanent:

echo 'export HSA_ENABLE_DXG_DETECTION=1' >> ~/.bashrc
source ~/.bashrc

Validation commands

After a wsl --shutdown from Windows and a fresh WSL session, check GPU detection with:

export HSA_ENABLE_DXG_DETECTION=1
lspci | grep VGA
rocminfo | grep -E "(Name|Marketing|gfx)"

Expected output looks like:

03:00.0 VGA compatible controller: Advanced Micro Devices [AMD/ATI] Radeon 8060S
Name:                    gfx1151
Marketing Name:          Radeon 8060S

Python test

He also tested the ROCm PyTorch stack in a Python 3.12 environment (create a virtual environment first). Make sure to install the ROCm-specific PyTorch wheels built for ROCm 7.2. You can find the latest ROCm wheels on the AMD ROCm repository. Here is how to install them:

wget https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.9.1%2Brocm7.2.0.lw.git7e1940d4-cp312-cp312-linux_x86_64.whl
wget https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.24.0%2Brocm7.2.0.gitb919bd0c-cp312-cp312-linux_x86_64.whl
wget https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.5.1%2Brocm7.2.0.gita272dfa8-cp312-cp312-linux_x86_64.whl
wget https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchaudio-2.9.0%2Brocm7.2.0.gite3c6ee2b-cp312-cp312-linux_x86_64.whl
pip3 uninstall -y torch torchvision triton torchaudio
pip3 install torch-2.9.1+rocm7.2.0.lw.git7e1940d4-cp312-cp312-linux_x86_64.whl torchvision-0.24.0+rocm7.2.0.gitb919bd0c-cp312-cp312-linux_x86_64.whl torchaudio-2.9.0+rocm7.2.0.gite3c6ee2b-cp312-cp312-linux_x86_64.whl triton-3.5.1+rocm7.2.0.gita272dfa8-cp312-cp312-linux_x86_64.whl

If you install into a non-virtual environment, you may need to add --break-system-packages.

Once installed, run a simple test to check whether PyTorch detects the ROCm GPU:

python3 -c "import torch; print(torch.cuda.is_available()); print(torch.cuda.get_device_name(0))"

Issues and limits

At the moment these are the main caveats in his setup:

rocm-smi is still not supported in WSL
Performance is good, but below native Linux (70-80% of native performance in some tests)
Driver and Windows updates may require rebuilding librocdxg

For production workloads, I still recommend validating on native Linux as well.

If you try this setup on different hardware or a different Windows/WSL stack, feel free to share your results and differences.

Introducing Chonkie.Net - Bringing RAG Ingestion to .NET

Wed, 18 Feb 2026 10:00:00 GMT

As part of a neverending journey in tinkering and exploring AI-assisted coding with GitHub Copilot, I'm excited to announce the preview release of Chonkie.Net, a faithful, high-performance port of the popular Python Chonkie library to the .NET ecosystem.

What is Chonkie.Net?

Chonkie.Net is a lightweight ingestion library designed to power fast, efficient, and robust Retrieval-Augmented Generation (RAG) pipelines in .NET applications. Whether you're building intelligent document processing systems, semantic search engines, or AI-powered applications, Chonkie.Net provides the foundational tools you need.

At its core, Chonkie.Net transforms raw text into perfectly-sized, meaningful chunks optimized for both language models and vector databases. But it's much more than simple text splitting: it's a complete ecosystem for document ingestion.

From Python to .NET

This project represents an interesting experiment in AI-assisted software engineering. What started as a personal curiosity, "How much can AI agents help when porting a complex Python library to C#?" — evolved into a fully-functional, production-ready library that I needed for several projects.

The journey was powered by GitHub Copilot and coding agents in Visual Studio and Visual Studio Code. Built during evenings and weekends over the previous months, the project represents approximately 15 days of full-time work. The results have been promising: high feature parity with the original Python library, idiomatic C# APIs, and a modular, dependency-light architecture.

My workflow was hybrid: I used iterative collaborative chats (where the assistant suggested code snippets and explained the Python implementation), agent mode for autonomous code generation (with manual adjustments for quality), and what's become known as "vibe coding" (providing high-level direction and letting the agent generate based on the Python reference implementation). This always followed a plan-implement-review loop with incremental steps, documentation, and tests. I'd review, suggest improvements, and iterate. Committing frequently and merging PRs at milestone endpoints kept the work organized as I progressed through each feature.

Some parts were more challenging than others, especially translating Python's dynamic typing and flexible data structures into C#'s statically-typed world. But overall, the experience was rewarding, educational, and fun. The result is a library that brings Chonkie's power to .NET developers while serving as a case study in how AI tools can augment and accelerate software development.

Key Features

What makes Chonkie.Net stand out in the .NET landscape for RAG applications? Here are the key highlights:

11 Specialized Chunkers – choose the right strategy for your data, from simple character-level splitting to intelligent recursive chunking and semantic extraction
Multiple Tokenizers – support for word-based, sentence-based, and transformer-based tokenization
7 Embedding Providers – seamless integration with OpenAI, Azure OpenAI, Google Gemini, Cohere, VoyageAI, Jina, and ONNX local models
9 Vector Database Integrations – write directly to Pinecone, Qdrant, Chroma, Weaviate, MongoDB, Pgvector, Elasticsearch, Milvus, and Turbopuffer
5 LLM Providers – integrate with OpenAI, Azure, Groq, Cerebras, and Gemini seamlessly
Local-First Design – run embeddings locally with ONNX Runtime, no cloud dependencies required
Type-Safe, Modern C# – built with C# 14 nullable reference types, dependency injection support, and idiomatic .NET patterns
High Performance – .NET 10 optimizations that match or exceed Python performance in many scenarios
Comprehensive Testing – 930+ unit and integration tests ensure reliability

Thanks to the amazing work of the original Chonkie team, the core algorithms and strategies have been faithfully implemented, while also taking advantage of .NET's strengths to provide a smooth developer experience.

Getting Started in 30 Seconds

Here's how simple it is to start chunking:

using Chonkie.Chunkers;
using Chonkie.Tokenizers;

var text = "Woah! Chonkie, the chunking library is so cool!";

var tokenizer = new WordTokenizer();
var chunker = new TokenChunker(tokenizer, chunkSize: 64, chunkOverlap: 8);

var chunks = chunker.Chunk(text);
foreach (var chunk in chunks)
{
    Console.WriteLine($"Chunk: {chunk.Text}");
    Console.WriteLine($"Tokens: {chunk.TokenCount}");
}

Building a Complete RAG Pipeline

For more advanced scenarios, use the fluent pipeline API to orchestrate the entire ingestion flow:

var sampleDir = "pipeline_docs";

var result = await FluentPipeline.Create()
    .FetchFrom(new FileFetcher(), sampleDir, "*.txt")
    .ProcessWith(new TextChef())
    .ChunkWith(new RecursiveChunker(
        tokenizer: new WordTokenizer(),
        chunkSize: 20))
    .RefineWith(new OverlapRefinery(minOverlap: 5))
    .ExportTo(new JsonPorter(), "pipeline_output.json")
    .RunAsync();

See the samples/ folder in the GitHub repository for many complete examples, including vector database integration and LLM interactions.

Project Status & Community

This is an early-stage, experimental project, a work in progress built in spare time for learning and exploration. While the core implementation is complete and the test suite is comprehensive, the project still needs real-world testing, bug fixing, and optimization before being considered fully stable for production use.

The good news? The foundation is solid, the architecture is clean, and the community is welcome. Whether you're interested in RAG systems, .NET AI tooling, or curious about the AI-assisted coding journey, there's plenty to explore.

Getting Involved

You can get started with Chonkie.Net in several ways:

Install from NuGet: dotnet add package Chonkie.Net (make sure to enable prerelease versions)
Clone and Build: Visit the GitHub repository to explore the source, contribute, or submit feedback
Read the Docs: Check out the comprehensive documentation and tutorials
Try the Samples: Browse practical examples in the samples/ folder to see Chonkie.Net in action

What's Next?

Future plans include deeper testing with production workloads, performance optimizations, expanded documentation with real-world scenarios, and community feedback. If you give it a try, please share your experience. Bug reports, feature requests, and contributions are always welcome.

If you're building your next RAG application in .NET, Chonkie.Net is here and ready to help.

Happy chunking! 🦛

AI Assistants at AI Heroes Torino, Dec 11th 2024

Wed, 30 Oct 2024 10:15:00 GMT

Busy weeks ahead in terms of events! I am pleased to announce that I will be a speaker, along with my colleague Morgana Lalli (R&D Data Scientist, at Deltatre), in the second edition of AI Heroes on December 11th, 2024, in presence in Turin and online.

AI Heroes is a full-day conference organized by Synesthesia with the support of numerous sponsors. The event offers a unique learning and networking opportunity for the community of developers, engineers, designers, and marketers working with artificial intelligence. Over 350 experts and enthusiasts are expected.

Check out the agenda and join us at 12:50 PM, we will have the session RAG-based AI Assistants: from demo to product, addressing real-world challenges: we will share our journey to bring an enterprise AI-based assistant into production, addressing the real-world complexities of development and deployment, sharing our experience with security, advanced ETL integrations, automation, and frontend/backend customizations. From a prototype based on Azure OpenAI Chat Playground (aka Bring-Your-Own-Data), we’ll walk you through how we moved to a custom architecture, optimized to meet specific user and business needs, using Microsoft technologies, including Azure AI Search, Semantic Kernel, and Kernel Memory.

To participate in the event, you can purchase tickets here, while for more information, please visit the official website.

Generative AI and Spatial Computing at WPC 2024 (Milan, Italy)

Wed, 30 Oct 2024 09:10:00 GMT

Also this year, I am pleased to invite you to the WPC 2024, the main Italian conference on Microsoft technologies. Organized in Milan (Italy) from November 26th to November 28th by OverNet Education. The list of speakers and sessions is ready and available online: as usual, also this year it is very rich, featuring numerous sessions on artificial intelligence, web development, app development, the latest updates on development tools, and many many other topics!

Together with my dear friend Clemente Giorio (Microsoft MVP, Intel Innovator), I will be speaker in the AI Development track in the afternoon, with the session Generative AI + Spatial Computing: technologies and applications from the future (presented in Italian). In the session we will see the latest developments in Deep Learning in this multidisciplinary field and explore techniques that allow us to reconstruct and analyze 3D scenes, objects, and people, discovering that uses and services once belonging to science fiction are now finally within our reach.

If you haven't registered yet, now is the time to do so!

Talking at Azure Day Torino 2024

Sun, 29 Sep 2024 12:30:00 GMT

As usual, after a summer break, communities and organizers are back at work at full speed, preparing new and amazing events where to share with you all the latest news about technologies and trends. Together with my colleague Morgana Lalli, we've been selected as speakers at the upcoming Azure Day Torino 2024 event, on October 19, 2024.

It is a free, full-day event organized by Microsoft, MSC Technology and supported by local Italian communities, focused on Microsoft technologies. Have a look at the agenda, and join us at the session AI Assistants, from click-deploy to production: we'll explore together the journey of taking a RAG-based business assistant from a demo to a fully operational solution, addressing the real complexities of development and implementation. We will share our experience on data pipelines, frontend and backend customizations, automation, monitoring, and security. Azure OpenAI Bring-Your-Own-Data allows for a quick demo setup: in this session, we will see how the system can evolve, transitioning to a customized architecture, still based on Microsoft technologies (including Azure AI Search, Semantic Kernel, and Kernel Memory) to meet specific user needs and business requirements.

Please, register here if you want to attend, and you can find more information on the official event website.

Don't miss it!

AI Conference 2024, Milan, Italy

Tue, 30 Apr 2024 14:12:00 GMT

Since 2019, AI Conference has been an event designed to bring Artificial Intelligence and Machine Learning themes to Italy. With Italian and international guests, this event will discuss everything CTOs, developers, ML Engineers, and data scientists need to address as the technical and ethical challenges of recent years. If you're interested, you're invited to the fourth edition organized in person by improove, along with some partners, sponsors, and local communities, in Milan (Talent Garden Calabiana), on June 17, 2024.

On this occasion, I will present, in Italian, a session that combines AI and sports: "Karate and AI: Revolutionizing Martial Arts with Deep Learning and Body Tracking", where I will share my initial explorations in applying the latest technologies, especially Deep Learning and Body Tracking models, to this ancient martial art. We will evaluate how adopting these solutions can not only enhance athletes' performance but also provide coaches with new and reliable analytical tools.

Registration is required to attend the event.

The agenda has been announced, featuring truly interesting topics and outstanding speakers. To stay updated on the event, visit the official event website.

In the meantime... share, and see you there!

Global Azure 2024 Torino, Italy

Wed, 10 Apr 2024 18:26:00 GMT

Global Azure is a free one-day event, organized by local communities around the world, focused on Microsoft Azure. If you are interested, I'd like to invite you to the event we're organizing in-presence in Turin, Italy, on April 20th, together with Azure Meetup Torino, Retelit, and other local communities.

The event is free, but registration is required.

The agenda is ready, there will be very interesting topics and amazing speakers. To stay up-to-date about the event organization, please check the official website for the Italian edition in Turin.

In the meanwhile... spread the word, and we'll see there!

Generative AI with XeDotNet friends

Tue, 02 Jan 2024 12:45:00 GMT

Happy New Year folks! Let's start the 2024 with an announcement: I've been invited by my friends of the XeDotNet Community to their first event of the year. Together with my good friend Clemente Giorio (Microsoft MVP, Intel Innovator), we'll present, in Italian, a long and practical session focused on our year-long explorations: Generative AI in practice: from the most famous services to DIY solutions. With some theory, lots of demos and hands-on code, together we will explore the use of generative AI for various use cases and we will see potential approaches for implementing custom services using open-source tools, frameworks, and models, leveraging on-premise and cloud computing power. It will take place on January 19th, 2024, online, starting from 9 PM (UTC+1).

To freely participate, you can register here.

AI Heroes, Dec 1st 2023

Wed, 15 Nov 2023 12:10:00 GMT

Busy weeks ahead in terms of events! I am pleased to announce that I will be a speaker, along with my colleague Luca Legora (R&D Technical PM, at Deltatre), in the first edition of AI Heroes on December 1, 2023, in Turin and online.

at 12:50 PM, we will have the session AI in Sports: from deep learning to generative AI: together, we will explore the world of AI applied to sports, where we will show how AI-based solutions are revolutionizing the viewing experience, fan engagement, coaching and referee support, and behind-the-scenes operations. Dive with us into the latest advancements in artificial intelligence technologies, such as deep learning, multimedia data analysis, generative AI, and information retrieval: we will explore cutting-edge innovations that are shaping the next generation of sports broadcasting and entertainment.
from 2:30 PM, we will have the technical workshop Generative AI Hands-On: from popular services to DIY solutions, with lots of demos and some code: together, we will explore the state-of-the-art in generative AI for various use cases: after providing an overview of the most renowned services offered by OpenAI, Microsoft, Midjourney, ElevenLabs, and other providers, we will present potential approaches for implementing similar services using open-source tools, frameworks, and models, leveraging on-premise and/or cloud computing power.

To participate in the event, you can purchase tickets here, while for more information, please visit the official website.

Global AI Community, Generative AI talks

Tue, 07 Nov 2023 22:00:00 GMT

Two amazing announcements today! Global AI Community organizers invited me and my good friend Clemente Giorio (Microsoft MVP, Intel Innovator) for the upcoming Global AI Notes weekly online appointment, on November 13th, 2023. !! UPDATE: Unfortunately, due to a flu, the event did not take place and will be rescheduled for a date yet to be determined.

The Global AI Notes is a weekly appointment, hosted by Global AI Community members, and delivered straight from the experts all around the world. From machine learning, robotics, big data, neural networks, and everything in between, it is a chat between speakers who share their latest discoveries and what's coming up next in the AI world. If you're a tech enthusiast or just curious, don't miss this exciting ride into the future of AI!

Join us at 10 AM (CET) for the session Generative AI Landscape: from popular services to open-source alternatives: we will explore the state-of-the-art of generative AI for various use cases. After providing an overview of the most renowned services such as ChatGPT, Midjourney, ElevenLabs, and RunwayML, we will show you potential approaches for implementing similar services on your own, by leveraging open-source tools, frameworks (such as Intel OpenVINO, ONNX Runtime, NVIDIA TensorRT), and models (such as SAM, Stable Diffusion, Whisper, LLama), while harnessing the computational power offered by your local workstation and/or Azure. You may already have seen a previous version of this session in Italian, but this is the first time we deliver it in English!

In addition, we've also been confirmed as speakers for another amazing event: the Global AI Conference 2023, always happening online, from all over the world, on December 12th, 2023.

The Global AI Conference is an all-day long event that brings together AI fans and experts from all over the world. It's put together by the Global AI Community, and it's a great opportunity to dive deep into the world of AI, so don't miss out!! AI developers, students, professional and everyone else passionate about AI should join this event!

We'll be presenting a brand new session: Generative AI Unleashed: crafting voices and music from text & images. While generative AI continues to evolve predominantly around text and imagery, the horizon is broadening with innovations in audio, synthesized voices, and music creation. This session provides an overview of the realm of generative AI for audio. Let's dive into hands-on examples, harnessing the power of open-source pre-trained models, and discover how the Microsoft ecosystem, among others, can be a powerful tool in this journey!

You're warmly welcomed to join us during our sessions: for registration details and how to attend the events, please refer to the official website. Don't miss those fantastic opportunities to learn more!

Excited to speak at AI Day 2023

Mon, 30 Oct 2023 12:10:00 GMT

As the fall season is now in full swing, communities and organizers are excited to share the latest news about technology trends and advancements. I am thrilled to announce that I will be speaker, remotely, together with my long time buddy Clemente Giorio (Microsoft MVP, Intel Innovator) at the upcoming AI Day 2023 event, on November 24th, 2023, in Rome (Italy) and online.

AI Day is a free, full-day event organized by DotNetCode Community with the support of Microsoft, other sponsors, and other Italian communities. The event is centered around artificial intelligence and its applications in .NET ecosystem. Check out the agenda, and join us at the session Generative AI Landscape: From Popular Services to DIY Solutions on Azure: we will explore the state-of-the-art of generative AI for various use cases. After providing an overview of the some of the most renowned services such as ChatGPT, ElevenLabs, and RunwayML, we will give you some hints about potential approaches for implementing similar services by leveraging open-source tools, frameworks, and models, while harnessing the computational power offered by Azure. You may already have seen a previous version of this session: the topic is the same, but demos are different!

To attend the event, please register here, and you can find more information on the official website.

Don't miss this fantastic opportunity!

Speaker at WPC 2023, Milan (Italy), November 28-30

Mon, 23 Oct 2023 21:30:00 GMT

I am pleased to invite you to the annual event WPC 2023, the main Italian conference on Microsoft technologies. Organized in Milan (Italy) from November 28th to November 30th by OverNet Education. The list of speakers and sessions is available online: this year, it is once again very rich, featuring numerous sessions on artificial intelligence, web development, app development, the latest updates on development tools, and many other topics!

Together with my dear friend Clemente Giorio (Microsoft MVP, Intel Innovator), we will be speakers in the Artificial Intelligence track, with the session Implementing Generative AI leveraging open-source models (in Italian). In the session, we will explore the state-of-the-art of generative AI for multiple use cases. After taking a look at ChatGPT, Midjourney, ElevenLabs, and RunwayML, we will see how to implement similar services leveraging tools, frameworks (Intel OpenVINO, ONNX Runtime, NVIDIA TensorRT), and some open-source models, harnessing the computational power of our workstation or Azure.

If you haven't registered yet, now is the time to do so!

Talking at Azure Day Torino 2023

Mon, 04 Sep 2023 16:35:00 GMT

After a summer break, communities and organizers are back at work at full speed, preparing new and amazing events where to share with you all the latest news about technologies and trends. Together with Clemente Giorio (Microsoft MVP, Intel Innovator), I've been invited as speaker at the upcoming Azure Day Torino 2023 event, on October 14, 2023.

It is a free, full-day event organized by Microsoft, MSC Technology and supported by local Italian communities, focused on Microsoft technologies. Have a look at the agenda, and join us at the session Generative AI Landscape: From Popular Services to DIY Solutions on Azure: we will explore the state-of-the-art of generative AI for various use cases. After providing an overview of the most renowned services such as ChatGPT, Midjourney, ElevenLabs, and RunwayML, we will delve into potential approaches for implementing similar services by leveraging open-source tools, frameworks, and models while harnessing the computational power offered by Azure.

Please, register here if you want to attend, and you can find more information on the official event website.

Don't miss it!

UGIdotNET Small Talks - Generative AI, a look into open-source models and toolkits

Tue, 02 May 2023 09:45:00 GMT

Together with the organizers and the UGIdotNET community, I'd like to invite you to the next live episode of Small Talks: Generative AI - A look into open-source models and toolkits, Thursday, May 4th, 5-6PM (CET). We will talk about generative AI, in Italian, and I will try to share my experience in some of its application scenarios. We will look at lesser-known models, frameworks, and tools that are open-source and, above all, that can be used to make the most of local resources on any PC, with or without a GPU, thanks to ONNX Runtime, Intel OpenVINO, and NVIDIA technologies.

Don't miss it!

NVIDIA GTC 2023

Mon, 06 Mar 2023 10:45:00 GMT

I’d like to invite you all to attend the NVIDIA GTC taking place March 20-23, 2023. It’s all virtual and free to attend: as you may know, GTC has become the must-attend event for researchers, developers, and business thought leaders in many fields and many industries, and this year promises even more exciting discoveries and interesting stuff to learn from.

You can explore more than 400 sessions at GTC, with talks that range from getting started in AI to the most advanced research in game development, cloud, the metaverse, and more. As always, the cornerstone of GTC is the keynote from NVIDIA CEO and founder, Jensen Huang. Watch it on March 21st, at 4 PM (CET) for a spotlight on NVIDIA's latest announcements and breakthroughs.

GTC is also a chance to connect with, and learn from, the experts. There are 50-minute sessions that will put you face-to-face with the people behind NVIDIA's ground-breaking products and research. Discover the latest trends in deep learning, accelerated computing, robotics, generative AI, metaverse applications, and so much more!

Whichever sessions you attend, you can look forward to learning from experts, connecting with colleagues, and exploring how NVIDIA can help you do your best work.

This year, Clemente Giorio (Microsoft MVP, Intel Innovator) and I have the opportunity to host a Watch Party session in Italian, together with Andrea Pilzer (NVIDIA Solution Architect): 3D Synthetic Data: Simplifying and Accelerating the Training of Vision AI Models for Industrial Workflows, on March 23rd, 1.30PM (CET).

Please register now for GTC, and feel free to share this link with your friends and/or communities for them to register as well.

Global AI Bootcamp 2023 Turin (Italy), March 4th

Fri, 24 Feb 2023 09:20:00 GMT

Also this year the Global AI Bootcamp returns, a free event organized by local communities around the world, focused on Artificial Intelligence with Microsoft technologies (but not only). As usual, me and Clemente Giorio (Microsoft MVP, Intel Innovator) are organizing the Italian edition, in presence (in Torino, Italy), on March 4th, 2023, from about 9AM to 6PM, with the help and support of Deltatre Innovation Lab, Global AI Community and some local communities, such as Torino Technologies Group (TTG), ItaliaDotNet, DotNet Campania e DotNetPodcast.

! UPDATE Feb 24th, 2023 !

The event is in-presence at Circolo del Design (Via S. Francesco da Paola, 17, 10123 Torino), kindly supported by our sponsors, Deltatre and Microsoft (Global AI Community).

Here's the agenda of the day:

Track 1 (in-presence, in Italian)

09:00 - 09:15 -> Opening
09:15 - 10:00 -> Keynote (Global AI community & Microsoft)
10:15 - 11:15 -> Azure AI and ONNX Runtime (Mauro Bennici - AI Architect & AI Ethicist)
11:20 - 12:20 -> Making AI optimization part of every developer's toolkit (Diego Fiori - Co-founder & CTO at Nebuly)
12:30 - 14:00 -> Lunch break
14:00 - 15:00 -> Costruire la città del futuro con Digital Twins e AI (Alessandro Festa - Sr. Product Manager at SmartCow)
15:10 - 16:10 -> Blackmirror - Creating your digital twin (Sebastiano Galazzo - Microsoft MVP, CTO at Synapsia-AI)
16:15 - 17:15 -> Azure OpenAI's API (Michele Aponte - Microsoft MVP, CTO at Ellycode and CEO at Blexin)
17:20 - 18:00 -> Hands-on OpenVino (Clemente Giorio & Gianni Rosa Gallina - Microsoft MVP, R&D Engineer at Deltatre)

Track 2 (Pre-recorded sessions by international speakers, in English):

14:00 - 15:00 -> Accelerating Pytorch with Intel Extension for Pytorch (IPEX) (Abhilash Majumder - Sr. HPC/Deep Learning Engineer at Intel)
15:10 - 16:10 -> Powering AI Capabilities with API Management and OpenAI API Integration (Bobur Umurzokov - Software Engineer)

We're waiting for you for this content-rich day.
Seats are limited, to attend is required to register at this link.

AI Day 2022 - Talking about Deep Learning Models optimization

Wed, 02 Nov 2022 11:50:00 GMT

I'd like to invite you to attend the AI Day 2022, back in presence in Rome (Italy) and also streamed online. It is a free full day of talks organized by .NET Code, on November 18. Together with Clemente Giorio, Microsoft MVP and colleague in Deltatre, I will co-present (remotely) the session Optimizing Deep Learning models: theory, tools & best-practices (in Italian), where we'll talk about theory and techniques to optimize Deep Learning models.

In this session we'll discuss the possible different layers of incremental improvements for deep learning models, from architecture review to dataset cleaning, and we'll share our experience leveraging Azure Machine Learning services and other open-source tools and frameworks.

The event is free, thanks to the sponsors, but registration is required. Agenda details can be found here.