We all recognise that 2023 was the year of AI, but more specifically of generative AI. We expect that there is still a lot of momentum in the generative AI space and everything related to LLMs. As such, the main trends that we foresee will focus mainly on generative AI and the advancement of new applications and techniques being developed in and on top of LLMs.

The next chapter in generative AI’s evolution we expect a shift towards more customised, intuitive, and multimodal gen AI applications. These trends, emerging from the collective efforts of leading tech companies and open-source communities, are set to redefine how businesses leverage AI for operational efficiency and consumer engagement.

1. Customised Gen. AI Applications: From LLM Agent to RAG

We are witnessing a significant shift in the field of generative AI. The previously dominant concept of ‘one size fits all’ is already changing fast with the rise of custom agents. Enterprises across various industries are recognizing the need for more tailored LLM solutions. Instead of relying on a single LLM, they are now adopting a strategy of leveraging multiple customised LLM agents. 

One of the key advancements in this area is the development of custom Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) capabilities. These advanced models have the ability to seamlessly connect and integrate data from diverse sources. By doing so, they are able to provide more accurate and informed responses to complex queries and challenges. The new RAG framework allows for the ever confident LLMs to actually admit when they are uncertain or don’t know something. (see the NVidia blog for a detailed explanation)

Combining these LLM agents together with the RAG framework allows for a whole new range of more autonomous operations and LLMs that can operate on knowledge they have internally. Instead of just having a chatbot for your customer service, you can now also develop a “chat agent” as well as an “service agent” that actually executes actions that come from the agent chatting to your customers. The advancement of this field is

2. Open Source LLM Models

Open-source LLM models are increasingly becoming an essential component of modern business strategies. Also the speed with which companies like Mistral are developing and releasing their models makes it hard to keep up. These models provide a significant advantage by enabling organisations to develop domain-specific AI applications quickly and efficiently without risking giving other organisations access to their proprietary data. With the ability to integrate these models with private or real-time data, businesses are experiencing a substantial reduction in costs versus relying heavily on the cost of using APIs of for example OpenAI’s GPT4. This blog post is a great example of when to choose open source vs. off-the-shelf.

This emerging trend is making LLMs more widely accessible and customisable, not only on cloud-based platforms but also within data centers and desktop environments. As a result, organisations of all sizes and industries can now leverage the power of LLMs to drive innovation and achieve their business objectives.

3. Gen. AI and Data Infrastructure: The need for Vector Databases

Generative AI and data infrastructure increasingly rely on vector databases for efficient and accurate operations. Vector databases store data as high-dimensional vectors, mathematical representations of features or attributes, allowing for fast and accurate similarity searches. This is crucial for generative AI applications, as it enables them to generate more relevant and coherent content by understanding the semantic or contextual meaning of the data. By using vector databases, generative AI can overcome challenges like inaccuracy or lack of factual consistency, enhancing their utility in various domains such as natural language processing and computer vision

But in 2024 the concept of “vector databases” is evolving, as it becomes apparent that most databases will eventually offer vector search capabilities. This trend includes various database types like graph, relational, document, and key-value databases. The distinction between traditional databases and specialised vector databases is blurring, with many incumbents integrating vector search to capture new workloads, particularly in retrieval augmented generation (RAG). This shift benefits customers by reducing complexity and costs, as it eliminates the need for separate vector databases and minimises data movement.

4. Gen. AI and Microservices

The proliferation of API endpoints and AI microservices, is transforming software development. Developers are now customising off-the-shelf AI models more efficiently, integrating these capabilities into applications without the overhead of maintaining underlying infrastructures. End-users benefit from more intuitive, responsive applications, adapted to their specific needs.

An example is what NVIDIA introduced at AWS re:invent: The NeMo Retriever, a generative AI microservice within the NeMo framework, designed to enhance AI applications by integrating large language models with enterprise data for more accurate responses. This service, available on AWS, offers advanced retrieval-augmented generation capabilities and supports a range of data types, allowing businesses to develop custom AI applications with improved accuracy and efficiency, including those in use by global leaders like Cadence, Dropbox, and SAP.

5. Multimodality in Gen. AI

Moving beyond traditional text-based interactions, multimodal Language and Learning Models (LLMs) are now enabling users to interact using not only text, but also speech and images. This advancement is of utmost importance as it allows for the delivery of contextually relevant responses, leading to a more immersive and dynamic user experience.

Companies at the forefront of this innovation, such as Meta, OpenAI and Google, are spearheading the development of multimodal AI technology. With its ability to process and analyse data in various formats, including PDFs, images, and graphs, multimodal AI has the potential to revolutionise numerous sectors by providing a more comprehensive understanding of information and facilitating more informed decision-making processes.

For example the newly released Gemini model by Google showcases amazing possibilities that are just around the corner. Also the video (although it has been somewhat post-processed and not 100% showing the current state of the art) is amazing to see. Definitely lots of exciting developments ahead!


The AI landscape of 2024 is marked by customisation, multimodal interactions, and enhanced data infrastructure capabilities. These trends are not just reshaping how businesses use generative AI but are also redefining the boundaries of what generative AI can achieve. As we embrace these advancements, the potential for innovation and efficiency in various sectors is immense.

The trends discussed here are just the tip of the iceberg in AI’s potential for businesses. For a more comprehensive guide, our whitepaper “Beyond the Hype: A Practical Guide for AI Integration in Business” delves deeper. It demystifies various AI types, from narrow to generative AI, and illustrates their real-world applications. The report also advises on when not to use AI and how to integrate it into a robust data strategy. It’s an essential resource for building data infrastructures that truly harness AI’s value.