👏 Bagrat Ter-Akopyan, Carmen Heger and team. Congrats and thanks for the nice technical write-up. It’s amazing to see how you continue to innovate on behalf of your customers. I love the outcome.

📖 What stands out for me from your technical report is your description of the why and how you moved away from your own, already very good, initial RAG implementation. Citing from the report:

“𝘖𝘯𝘦 𝘰𝘧 𝘰𝘶𝘳 𝘭𝘦𝘢𝘳𝘯𝘪𝘯𝘨𝘴 𝘧𝘳𝘰𝘮 𝘵𝘩𝘦 𝘭𝘢𝘴𝘵 𝘣𝘪𝘨 𝘙𝘈𝘎 𝘱𝘳𝘰𝘫𝘦𝘤𝘵 𝘸𝘢𝘴 𝘵𝘩𝘢𝘵 𝘸𝘦 𝘯𝘦𝘦𝘥𝘦𝘥 𝘵𝘰 𝘴𝘱𝘦𝘦𝘥 𝘶𝘱 𝘰𝘶𝘳 𝘦𝘹𝘱𝘦𝘳𝘪𝘮𝘦𝘯𝘵𝘢𝘵𝘪𝘰𝘯 𝘤𝘺𝘤𝘭𝘦𝘴. 𝘖𝘯 𝘵𝘩𝘦 𝘵𝘦𝘤𝘩𝘯𝘪𝘤𝘢𝘭 𝘴𝘪𝘥𝘦, 𝘵𝘩𝘢𝘵 𝘮𝘦𝘢𝘯𝘵 𝘶𝘴𝘪𝘯𝘨 𝘮𝘰𝘳𝘦 𝘰𝘧𝘧-𝘵𝘩𝘦-𝘴𝘩𝘦𝘭𝘧 𝘴𝘰𝘭𝘶𝘵𝘪𝘰𝘯𝘴 𝘧𝘰𝘳 𝘙𝘈𝘎 𝘢𝘳𝘤𝘩𝘪𝘵𝘦𝘤𝘵𝘶𝘳𝘦𝘴 𝘵𝘩𝘢𝘵 𝘩𝘢𝘷𝘦 𝘣𝘦𝘤𝘰𝘮𝘦 𝘢𝘷𝘢𝘪𝘭𝘢𝘣𝘭𝘦 𝘢𝘯𝘥 𝘳𝘦-𝘶𝘴𝘪𝘯𝘨 𝘰𝘶𝘳 𝘰𝘸𝘯 𝘦𝘹𝘪𝘴𝘵𝘪𝘯𝘨 𝘴𝘰𝘭𝘶𝘵𝘪𝘰𝘯𝘴.” and “𝘛𝘰 𝘮𝘢𝘬𝘦 𝘵𝘩𝘦 𝘥𝘰𝘤𝘶𝘮𝘦𝘯𝘵𝘴 𝘴𝘦𝘢𝘳𝘤𝘩𝘢𝘣𝘭𝘦 𝘧𝘰𝘳 𝘰𝘶𝘳 𝘙𝘈𝘎 𝘴𝘺𝘴𝘵𝘦𝘮, 𝘸𝘦 𝘶𝘴𝘦𝘥 𝘈𝘞𝘚 𝘬𝘯𝘰𝘸𝘭𝘦𝘥𝘨𝘦 𝘣𝘢𝘴𝘦𝘴. 𝘛𝘩𝘪𝘴 𝘨𝘢𝘷𝘦 𝘶𝘴 𝘴𝘭𝘪𝘨𝘩𝘵𝘭𝘺 𝘧𝘦𝘸𝘦𝘳 𝘤𝘰𝘯𝘧𝘪𝘨𝘶𝘳𝘢𝘵𝘪𝘰𝘯 𝘰𝘱𝘵𝘪𝘰𝘯𝘴 𝘵𝘩𝘢𝘯 𝘢 𝘤𝘶𝘴𝘵𝘰𝘮 𝘣𝘶𝘪𝘭𝘥 𝘦𝘢𝘳𝘭𝘪𝘦𝘳 𝘴𝘰𝘭𝘶𝘵𝘪𝘰𝘯 𝘣𝘶𝘵 𝘴𝘱𝘦𝘥 𝘶𝘱 𝘪𝘯𝘨𝘦𝘴𝘵𝘪𝘰𝘯 𝘣𝘺 𝘴𝘦𝘷𝘦𝘳𝘢𝘭 𝘮𝘢𝘨𝘯𝘪𝘵𝘶𝘥𝘦𝘴 𝘸𝘩𝘪𝘤𝘩 𝘢𝘭𝘭𝘰𝘸𝘦𝘥 𝘶𝘴 𝘵𝘰 𝘪𝘵𝘦𝘳𝘢𝘵𝘦 𝘱𝘳𝘦-𝘱𝘳𝘰𝘤𝘦𝘴𝘴𝘪𝘯𝘨 𝘮𝘶𝘤𝘩 𝘧𝘢𝘴𝘵𝘦𝘳.”

📢 If you are on the hunt for an image 𝗔𝗡𝗗 𝗩𝗜𝗗𝗘𝗢 segmentation model, which is open and you can deploy on your own, have a look at the just released 𝗦𝗲𝗴𝗺𝗲𝗻𝘁 𝗔𝗻𝘆𝘁𝗵𝗶𝗻𝗴 𝗠𝗼𝗱𝗲𝗹 𝟮 (𝗦𝗔𝗠 𝟮). The model capabilities can be nicely experienced in Meta’s Demo). Read more about the announcement at their announcement page.

👷‍♀️If you are looking into deploying the model to build your own application on AWS, 𝗔𝗺𝗮𝘇𝗼𝗻 𝗦𝗮𝗴𝗲𝗠𝗮𝗸𝗲𝗿 is a very good alternative for you. Quoting from Meta’s announcement website:

⏰ Early start and back to Zürich Airport! 🛫 Some disturbances grounded me here last night, but I’m looking back to yesterday’s amazing Generative AI hackathon with a smile 😊.

☕ The participants were very engaged and built impressive PoCs in no time. My favorite feedback was: “It’s amazingly easy to create an 80% solution in almost no time”, “Based on the learning I built a 2nd use case in just 10 more minutes and got amazing results”, and “It’s not just the Generative AI models, but the platform that allowed me to build RAG-based solutions so easily”. 🙌

What a great evening yesterday at the AWS User Group Munich. The quote of the evening for me, yeah I might be a little biased 😉 , was :

“There are multiple model launches in Amazon Bedrock. Launching new models in Bedrock became like new minor versions numbers for databases or other services. Hard to keep up with the pace”.

Reflecting on this, I think that is a very true statement. It underlines the pace of innovation in this space and the necessity of choice at your hands if you want to build #GenAI based application for your customers. Hence the need of a service like #AmazonBedrock. Also featured the Bedrock Converse API, which makes switching models even easier.

How do you choose the foundation model for your Generative AI App — like your car?

Just published a new blog post on medium:

😎 How do you choose the right foundation model for your Generative AI app? 💡 It’s like picking the perfect car!  🚗 Just like car buyers care more about features, price, and user experience, Generative AI app users prioritize functionality, cost, and ease of use over the specific model under the hood.  👷‍♂️ Building a successful Generative AI app requires a well-architected, cost-effective solution that meets customer needs, not just chasing the latest, most powerful model.  🧠 Models are crucial, but they’re just one component. A robust architecture with components like data management, model orchestration, and user interfaces is essential.  💰 Cost-effectiveness matters. Choose the “Goldilocks” model that’s good enough for your use case, not overpowered (and overpriced).  ⏱️ Models evolve rapidly. Your app architecture and processes must allow seamless model updates or replacements to stay competitive.  🧩 Embrace model composability. Use different models for different tasks within your app for optimal cost and performance.  🛣️ Just like cars, Generative AI apps aren’t one-size-fits-all. Tailor your solution to your specific use case and customer needs.  🚀 Don’t get caught up in the hype. Work backward from customer problems to build fantastic, innovative solutions with Generative AI.  🎉 Enjoy the ride and have fun building your Generative AI app! It’s an exciting journey ahead.

What a very looong weekend in Stockholm. Loved every second of it. Started of with spectating the #StockholmMarathon. Would have loved to join, but no last minute (literally) bibs available. Only learned about it in the hour before the start. Still had a very good time.

We started the week with a Prompt Engineering on #AmazonBedrock workshop leading participant through the implementation of a marketing use case. Was great to see Oliver Möller, Tobias Nitzsche and Chakkree Tipsupa in joined action. The workshop has been good received and AWS office’s in Stockholm have been a very welcoming place for all the customers.

Prompt Engineering is possibly the single most valuable skill you have to master if you want to get to production with your Generative AI based application.

As many customers are working hard towards bringing their ideas and proof of concepts to realisations in 2024, this topic hits the keynote stage of the #AWSSummit in Stockholm.

We bring:

• 🛠️a toolbox of proven tools to build reliable prompts
• 👷a mechanism to create reliable results, turning trial&error into engineering
• 📊a ton of learnings from idealo internet GmbH ‘s journey into production

Looking forward to meet you there 😊

The ones who joined Philipp and me in our session at the #AWSSUMMIT in Berlin last week already got a preview of the blog post as we run it as a demo. Nice that it is now published and you all can get hands on it. Kudos!

AWS Inferentia2 is a great way to optimize the inference part of your (gen)AI workloads on AWS and the blog post helps you to dive straight into deploying a LLM (in this case Meta’s Llama 3) model. But it is not “just” Llama 3. From Hugging Face recent blog post: “Enabling over 100,000 models on AWS Inferentia2 with Amazon SageMaker” - https://lnkd.in/ePYb6TFs. So there is a good chance that you can benefit from AWS Inferentia 2 today :)

💡On my way back from a customer workshop on “Prompt Engineering” in Den Haag in the Netherlands. Good to connect with nature and an upcoming storm and very interesting to learn from customers about their experiences with GenAI. Still on the mission of turning authoring a prompt from a pure art form to more of an engineering approach. A Test driven, automated engineering approach for creating prompts has well resonated with the participants of the workshop - can’t wait to see that implemented. Big kudos to the participants who turned a potential boring presentation in an interactive exchange of ideas. Loved it!