Summary of the NDC Conference 2024

Artificial Intelligence

As you could expect, AI was a hot topic with a lot of sessions about it. This area is still very new and rapidly changing. In on session there was a poll on how many of the participants that had actually released functionality that used AI in production, and only 5-10% raised their hands.

Several sessions discussed how tokenization of language works in reality, and it was interesting to understand how tokenization can differ quite dramatically depending on the input language and the language that the Large Language Model, or LLM for short, has been mostly trained with. For example, a Norwegian team has compared using content in English and Norwegian with an Open AI language model (mostly trained in English) and a model mostly trained in Norwegian, and they saw that tokenization matched words when having input in the language model that the LLM was trained in, while the same model would create a less obvious tokenization when working with input in another language. Also, the number of tokens created would be much higher in less familiar languages. This has a direct effect on the cost of using the AI services since this is usually based on tokens.

Other sessions talked about the fragility of LLM:s and how they are very sensitive to input. For instance, the prompts “Create a quiz about [topic] in Swedish” and “Create a quiz in Swedish about [topic]” will deliver totally different results.

Also, there is instability in responses, and even if you are very explicit with your prompt, you’ll get strange/wrong answers now and then. You need to validate the response and have logic to handle these cases, like asking the LLM again. LLMs can also return empty responses, so testing and applying logic for validating data and how the system behaves is crucial.

Data and Energy

Sustainability was also a new topic with both dedicated talks about the subject as well as mentions in other sessions including keynotes. It’s interesting to hear how cloud providers seem to focus on energy generation themselves in a world with rapidly increasing needs for this fueled by the expansion of AI.

• Microsoft has hired people with nuclear expertise.
• Google is new running Geothermal Energy to run their datacenters
• There were sessions on how to be able to reduce emissions of an application by reducing energy usage, using energy that has low carbon emissions based on time and location.
• Microsoft showed how improved scaling for Azure Functions makes it easier to not have to pre-allocated unnecessary resources by dynamically rapidly changing resources demanding on demand.

Epinova recently held a seminar about the subject and it’s possible to see the recording here (In Swedish): 

Accessibility

There were a few topics regarding accessibility. Even if this topic is not new in the software development industry, the European Accessibility Act (EAA) that comes in effect in June 2025 makes this a relevant topic. Here are some highlights in the sessions:

• 15% of people globally (1.2 billion) have a permanent disability. $13 trillion income disposable for these people.
• CSS Nestings are now included in CSS. According to one of the speakers, this is one of the last reasons to use a CSS preprocessor.
• Lots of supports to do interaction that previously required JavaScript is now built into HTML and CSS.

If you want to understand more about EAA, we’ve published a blog post about the topic.

Microsoft Aspire

Scott Hunter talked about Microsoft Aspire that was launched on this year’s Build conference. Microsoft Aspire tries to simplify the process of building cloud native applications by automatically connecting resources to each other given the current environment, which means that you don’t have to spend so much time on handling this yourself. This makes it much simpler to push a locally developed application into the cloud.
It includes features like the Aspire Dashboard for real-time monitoring, application orchestration, and pre-configured components for resilience and observability. The stack supports seamless service discovery and deployment to environments like Azure and Kubernetes.
In Scott Hunters session on day two he also talked about improvements in scaling for Azure functions as well as other improvements in developer support for these to make them work more like web apps. You now have access to stuff like ILogger, modelbinding for parameters and more. One positive benefit with these improvements in scaling is that it’s possible to turn down the minimum number of instances for your function that will reduce energy consumption as mentioned earlier in the blog post.

Improvements to Entity Framework Core

One session went through improvements in the latest versions of Entity Framework Core, with version 8 that was release late last year. Here’s a few highlights:

JSON Column

EF Core now supports JSON columns, allowing you to store and query JSON data directly within your database. This feature is particularly useful for applications that require flexible data models or need to store semi-structured data. With JSON columns, you can efficiently store and retrieve JSON objects, perform complex queries using LINQ, and leverage the full power of relational databases while maintaining the flexibility of NoSQL-like data structures.

Hierarchical ID

EF Core also introduces support for hierarchical IDs, making it easier to model and query hierarchical data structures such as organizational charts, product categories, or nested comments. This feature simplifies the handling of tree-like structures in your database, enabling more efficient data retrieval and manipulation. By using hierarchical IDs, you can improve the performance and maintainability of your applications that rely on nested data relationships.

Many-to-Many Relationships

EF Core now provides robust support for many-to-many relationships without requiring a join entity class. This enhancement simplifies the modeling of many-to-many relationships between entities, such as students and courses or products and categories. With this feature, you can directly define many-to-many relationships in your entity classes, making your data models more intuitive and easier to work with.

Other Miscellaneous things that we heard about

Azure Logic Apps is a technology that makes it easy to create an run automated workflows with little or no code. There are lots of prebuilt apps that you can use, for instance with Teams. Logic apps is an alternative to using Power Automate (which is built on top of logic apps).  At Epinova we've been using Logic Apps to send status alerts for solutions hosted in Azure to Slack. You can read a comparison between these two technologies here.

Of course, there would be no tech conference without new acronyms. One that I remember is RAG – Retrieval-Augmented Generation – a technique for enhancing the accuracy and reliability of generative AI models. 

Semantic Kernel is an open-source development kit to simplify integrating AI services into your applications.

There are improvements in Application Insights that makes it possible to see potential code optimizations based on code running in your application. You can find it under Application Insights -> Performance -> Code optimizations. If you have Copilot, you can even connect this directly to Application Insights to get a suggestion of the code to change which makes this a very effective process for optimizing your applications.

Azure Load Testing is a tool for ensuring the performance and reliability of your Azure Functions under varying loads. It allows you to simulate high traffic scenarios and identify potential bottlenecks in your application. It fires up multiple server instances within seconds and can be monitored in App Insights Live Metrics.