deepset Cloud: An LLM Platform for AI Teams

Our enterprise-grade platform guides and empowers cross-functional teams throughout the AI product lifecycle

Large language models (LLMs) are the breakthrough technology of our time, and companies are racing to integrate these computational marvels into their products. deepset Cloud drastically reduces the overhead associated with building a production-ready application powered by LLMs, freeing teams to focus on the product itself. 

In this blog post, we’ll take you on a tour of our platform’s core principles to help you decide whether your team would benefit from a cloud-based platform for building with LLMs.

The promise of large language models

The amazing triumph of LLMs following the release of ChatGPT in November 2022 comes down to two main factors: their ability to understand and reproduce language and knowledge, and their incredible versatility. LLMs can be used to generate content, chat with customers, and perform other complex tasks previously reserved for humans alone, such as generating SQL queries and reviewing code.

We’re nowhere near the limits of what these models can accomplish. This makes it all the more exciting to be at the forefront of building with LLMs. But it also brings with it some unique challenges.

Your project, carried by the cloud

When it comes to AI-powered products, your team of data scientists, backend engineers, and product designers should focus on the quality and usefulness of the application they’re building. However, that’s only possible if the heavy lifting  – such as the scaling of compute resources and managing multiple pipelines – is taken care of.

A schematic cloud with three pairs of differently coloured figures grouped around it, there's a blue pair labeled "Data science" on the left, a red pair "labeled "Product" on the right, and a green pair labeled "Engineering" at the bottom. Inside the cloud, there are numbered items: 1. "Manage and index data at scale," and a diagram of documents which enter a pipeline, from which an arrow points to a database icon. Beneath the pipeline, there are two arrows in a circular arrangement. From here, an arrow points to 2. "Design and deploy prototypes": Two semi-transparent pipelines are connected to symbols outside of the cloud labeled "Hugging Face", "OpenAI", "Cohere", "Anthropic". This section is labeled "Connect to different LLM providers". From point 2, an arrow points to 3: "Deploy to production" over a combined icon of a UI, a gear symbol and a wrench. Point 4: "Manage compute resources" above a symbol of lines and nodes, with an arrow to 5: "Monitor performance" above a diagram of a graph. From here, an arrow leads back to point 2 (Design and deploy prototypes). Finally, point 2 connects to an icon of a browser in the upper right corner of the image with thumbs-up/ thumbs-down symbols. Next to a small group of figures, it says "Demo to stakeholders". Next to a slightly larger group of figures it says "Test with users". From point 3 (Deploy to production), another arrow points to that same browser icon. The arrow is labeled "Serve LLM features at scale".

Enter deepset Cloud. Our LLM platform lets you design, deploy, monitor, and evaluate your AI application in one clean and straightforward UI (you can watch the demo video here). It also manages compute resources seamlessly in the background, while functioning as a unified environment for cross-functional teams to create the best product possible: powered by LLMs, and by your company’s data.

deepset Cloud’s core design principles

Let's take a closer look at the key pillars of our platform to see exactly how it helps organizations – from legal and financial institutions to online news media – build LLM-powered products in a fast, clean, and easy-to-use way.

Hedge your bets in the high-paced LLM space

LLMs use real-world textual training data as a basis to learn human-like reasoning. New LLMs are appearing all the time: smaller models that require less processing power, customizable open-source LLMs, or models that can use huge context windows as the basis for their output.

In such a fast-moving field, it is essential to stay on top of new developments – and equally important not to get locked into any one model or model vendor. That is why deepset Cloud is built with composability in mind: our proven pipeline paradigm allows teams to build systems with the flexibility to run different models independently. This modular approach also means that you can easily adapt your system to changing user needs without having to redesign it from scratch.

Build with the user experience in mind

Your product is only useful if it solves your users' pain points. That's why we encourage developers to think about user needs from day one. And why we built deepset Cloud with easy prototyping and feedback in mind: within the UI, you can set up multiple versions of your system and test them directly with real users. Quantitative metrics let you compare different setups side-by-side at a glance.

Note that there’s no need to build anything yourself: every prototype comes with a user interface and a link, as well as highly intuitive feedback features. Your users will love it!

Accept only the highest level of security 

When it comes to sensitive proprietary data, you can not be too careful. That's why deepset Cloud meets the highest security standards and has passed the SOC 2 Type II cloud provider security framework. As a result, our customers sleep well at night knowing that their data is stored securely and in compliance with the latest security protocols. 

For customers who are concerned about sending their proprietary data to companies like OpenAI – within the LLM prompt, or even to fine-tune a model  – deepset Cloud can connect to open source models on AWS SageMaker.

Align your stakeholders for LLM adoption

In traditional data science workflows, there can be a lack of transparency. Teams may spend months or even years on designing the perfect system, without ever testing it with real users or showing it to stakeholders. 

However, experience shows that a practice of rapid iterations is much more likely to turn your project into a success story. That is why deepset Cloud allows you to instantly demo your preliminary product to all types of stakeholders, regardless of their technical background.

Follow best LLMOps practices

As impressive as LLMs may be – within a production-ready application, they are just one element of a huge and intricate software system with lots of moving pieces. So when building with LLMs, it’s important to stick to the best practices long established by software developers:

  • Elasticity: LLM-powered pipelines deployed with deepset Cloud scale with your user base. The cloud-native architecture seamlessly adjusts to increased usage and saves you money during low-traffic times.
  • Managing data at scale: deepset Cloud takes care of the entire data lifecycle: from the ingestion of gigabytes of text to data processing and vectorization, to storage in a vector database. This can be a one-time event as well as a continuous process.
  • Evaluation: deepset Cloud places a huge emphasis on transparency. Find the best possible configuration using the built-in quantitative and qualitative evaluation tools.
  • Monitoring: deepset Cloud offers extensive observability features. It lets you monitor your pipelines’ output in real time, allowing you to keep an eye on your system’s behavior and take action when needed.
  • Agility: We respond to new challenges quickly. For instance, we were among the first to develop a successful mechanism for combatting LLM hallucinations – which we promptly integrated into deepset Cloud.

Why choose deepset Cloud?

deepset has been empowering enterprises to apply NLP and LLMs for over five years. Generative AI, which has evolved at lightning speed over the past few months, has long been on our radar – along with other groundbreaking technologies like information extraction and semantic search.  

Haystack, our open-source framework for building standalone applications with LLMs, is a comprehensive toolbox beloved by AI experts and newcomers alike. deepset Cloud leverages Haystack technology, and has inherited its composable and flexible philosophy. 

Through its fully-fledged cloud-based development and inference platform for enterprise AI teams, deepset Cloud adds ease of use, quick prototyping-feedback cycles, and a powerful, scalable backend architecture.

If you want to learn more about what deepset Cloud can do for you, contact us today.