The AI Alliance

⚛️ We're excited to be co-hosting LeMaterial Reading Group in 2026. The intersection of AI, materials science, and chemistry is exciting and moving at lightning speed. If you work in this space, you need to be part of this conversation. Check the details and add yourself to a session that works for your schedule. Know someone doing relevant work? Bring them along. The best insights come from people actively building in this field. Co-hosted with Entalpic Luis Pinto

Join us for talks about The Past, Present, and Future of AI Agents. Our speakers include: #1 Verifiably Safe Agentic Compute AI-Native Coding by Erik Meijer, Senior Fellow at Normal Computing #2 Tools! A history of agents doing stuff by Evan Tahler, Arcade.dev #3 Agentic Commerce by Ankit Agarwal, Skyfire #4 Panel: past, present, and future of AI Agents Erik Meijer, Evan Tahler, Vincent Koc, …

🧠 Why domain-specific agents matter Most professionals do not need “more AI.” They need AI that actually understands their work. That is what stood out at the Berlin AI Hack Night for Finance. Teams were not building generic assistants. They were building agents designed for financial workflows, with awareness of structure, rules, risk, and real operational constraints. For builders, that meant agents with domain-aware logic, task-focused models, and clear boundaries between automation and escalation. General agents are impressive. Domain-specific agents are deployable. This is the direction the AI Alliance is amplifying across finance, science, materials, climate, and industrial systems. 📸 These moments from Berlin capture the shift toward AI that works in the real world. Ready to get involved? https://lnkd.in/edN4jGrc cognee distil labs Selim Nowicki Jacek Golebiowski Atanas Iliev Agata Ferretti

🚀 Berlin didn’t just hack. It built with intent. This week, the AI Alliance AI Hack Night for Finance brought together AI hackers, builders, and agent dreamers for an evening focused on one thing: 👉 Domain-specific agents for real financial workflows. This was not a generic “build an agent” sprint. Teams designed agents that understand finance, including invoices, procurement, anomalies, structured data, risk, and context, and turned those ideas into working systems in hours. 🧠 That focus matters. Because agents only become useful when they are grounded in domain knowledge, domain data, and domain constraints. 🔥 What got built -Finance-aware agents with memory and inference -Automated invoice triage and procurement workflows -Anomaly detection and finance-specific Q&A -Small-model, task-focused systems designed to deploy, not demo 🏆 Hack Night Winners 🥇 Finance Guardian Agents Glib Gavryliuk Anton I. 🥈 Procurement Panda David Kaufmann Manuel Muellerschoen Niv Adam 🥉 Finance GPT Aleksandr Rasskazchikov Sara Ungureanu Georgi Georgiev Maria Daniela Pitulice Anna Wawrzyniec 🍕 Fueled by food, curiosity, and collaboration 🤖 Built by teams who showed up to ship Huge thanks to cognee and distil labs for supporting builders pushing agentic finance forward, and to everyone who came solo, teamed up fast, and built something real. 👀 Missed Berlin? More domain-focused hack nights are coming. 💥 This is where domain-specific agent builders meet. Follow The AI Alliance to build what’s next. Agata Ferretti Atanas Iliev

🚀 AI hackers, builders, and agent dreamers — this one’s for you! 🔥 The AI Alliance presents - AI Hack Night for Finance 📍 Dec 10 · 17:00–22:00 · Berlin 🎟️ Free · Limited spots → lu.ma/4bq1g9v8 Build smart agents with memory, inference & small models. Automate real finance tasks: invoice triage, anomaly detection, Q&A. 💸 Win up to €1000. Hack. Win. Connect. 👥 Teams of 3 — come solo or squad up 🍕 Food, prizes, energy 🤖 Brought to you by The AI Alliance 🧠 Sponsored by cognee & distil labs 💥 This is where Berlin’s agent builders meet. Will your agent rise? #TechEvents #AIHackNight #LLM #AIAlliance #BerlinTech #OpenSource #FinanceAI #Cognee #DistilLabs #AICommunity #AutonomousAgents

⚛️ 𝐃𝐞𝐜 𝟏𝟎𝐭𝐡 𝐋𝐞𝐌𝐚𝐭𝐞𝐫𝐢𝐚𝐥 𝐑𝐞𝐚𝐝𝐢𝐧𝐠 𝐆𝐫𝐨𝐮𝐩 Join us next week as we kick off the reading group featuring "Graph Diffusion Transformers are In-Context Molecular Designers" by Gang Liu 📅 𝐖𝐞𝐝𝐧𝐞𝐬𝐝𝐚𝐲, 𝐃𝐞𝐜 𝟏𝟎𝐭𝐡 𝐚𝐭 𝟖:𝟑𝟎𝐀𝐌 𝐏𝐓 / 𝟏𝟏:𝟑𝟎𝐀𝐌 𝐄𝐒𝐓 / 𝟓:𝟑𝟎𝐏𝐌 𝐂𝐄𝐒𝐓 𝐀𝐝𝐝 𝐝𝐢𝐫𝐞𝐜𝐭𝐥𝐲 𝐭𝐨 𝐆𝐨𝐨𝐠𝐥𝐞 𝐂𝐚𝐥𝐞𝐧𝐝𝐚𝐫: https://lnkd.in/edkCgq35 𝐃𝐞𝐬𝐜𝐫𝐢𝐩𝐭𝐢𝐨𝐧: In-context learning has shown limited success in molecular design, where labeled data remain scarce. We introduce demonstration-conditioned diffusion models (DemoDiff), defining contexts using molecule-score examples to guide a denoising Transformer. For pretraining, we develop a molecular tokenizer with Node Pair Encoding requiring 5.5× fewer nodes. Pretrained on millions of context tasks, our 0.7-billion-parameter model matches or surpasses language models 100-1000× larger, achieving an average rank of 3.63. These results position DemoDiff as a molecular foundation model for in-context design. 𝐏𝐚𝐩𝐞𝐫 𝐥𝐢𝐧𝐤: https://lnkd.in/ePEpq52F 𝐘𝐨𝐮 𝐜𝐚𝐧 𝐟𝐢𝐧𝐝 𝐩𝐫𝐞𝐯𝐢𝐨𝐮𝐬 𝐭𝐚𝐥𝐤𝐬 𝐡𝐞𝐫𝐞: https://lnkd.in/eSgPK5hU Co-hosted by Entalpic & The AI Alliance

🚀 Special Edition Office Hours: PyTorch 2025 Joseph Spisak (Product Director, Meta Super Intelligence Labs) is joining us for a special Office Hours this Friday to announce what's next for PyTorch. What's dropping: → OpenM: New collaboration with Hugging Face to define the standard for RL environments → TorchComms: Single interface for fault-tolerant communication collectives across diverse hardware → Monarch: Distributed training controller that treats GPUs as meshes. Multi-data center pre-training. Fully async RL. → Open-sourced internal stacks: Production-grade kernels, RL post-training, and agent-building methods → PyTorch Foundation news: Ray joins as a hosted project. Dell Technologies joins the board. No recording. Live Q&A with Joe. 📅 Friday, Dec 5 | 10 AM PST / 1 PM EST ⏱️ 60 min 🎟️ Free Register: https://lnkd.in/eNvbyGy3 #PyTorch #OpenSource #ReinforcementLearning #DistributedTraining #MachineLearning Andrea Greggo Tim Bonnemann

The MPDL, Max Planck Society has joined the AI Alliance. They specialize in trusted, open research data, expertise that directly supports our work on open, secure, and responsible AI. Their first contribution: access to bloxberg, a decentralized blockchain for time-stamping, certifying, and verifying research data. It's tamper-proof and transparent. Exactly what trustworthy AI development requires. MPDL joins a global network of researchers, developers, and institutions pushing open innovation forward. Welcome to the AI Alliance!

🍎 New York showed up big. Last month, the AI Alliance community packed New York University for two back-to-back events focused on open, agentic AI: 🔧 MCP-Agent Workshop  🤝 AI Agent Unconference The day began with a hands-on deep dive into MCP-Agent, an open-source framework for building composable agents using the Model Context Protocol. Special thanks to Sarmad Qadri of LastMile AI for sharing how to build a Deep Research Agent with MCP-Agent and leading developers through practical workflows. In the afternoon, the community transitioned into the AI Agent Unconference, where attendees co-created the agenda, formed spontaneous discussion groups, and explored real challenges in agent development—from evaluation and safety to governance and emerging use cases. A big thank you to our partners NYU Tandon School of Engineering, IBM, and the developers, researchers, founders, and builders who participated. Your collaboration is what continues to push open innovation forward. More meetups, workshops, and unconferences are on the way. If you’re building in open-source AI, we’d love to connect with you. Dave Nielsen Greg Spektor Rui Li #AIAlliance #AIAgentMeetup #MCPAgent #OpenSourceAI #AgenticAI #Developers #NYU #IBM