Jasper AI Orchestrates Multiple LLMs and Accelerates Innovation

TL;DR

Jasper AI is giving marketers next-generation tools to accelerate their business dramatically, from customized content creation to distribution to measuring impact. Jasper is positioned to be the generative AI company for marketing, giving marketers the blueprint for the future of applied AI.

Using Anyscale, Jasper AI is able to build, deploy, and manage multiple LLMs to automate a wide range of marketing tasks. They deliver a best-in-class AI-powered user experience while reducing cloud costs and accelerating development with the Anyscale Platform.

Overview

The generative AI transformation of marketing is already well underway. By 2025, researchers expect 30% of outbound marketing messages from large brands to be synthetically generated by AI.

Jasper is an AI copilot for marketing teams that want better outcomes. Jasper securely trains on a customer’s brand and strategy, accelerates the production of content (e.g. email, articles, social media posts), images, and campaigns, and helps marketers measure and optimize for performance, all via an end-to-end copilot. 

Jasper has been recognized as “one of America's fastest-growing private companies” by Inc. 5000 and was among the fastest companies to reach 100 million users. Customers include Morningstar, Anthropologie, SentinelOne, ZoomInfo, and more.

The Challenge

Delivering on this incredible vision presented a number of technology challenges for the product team at Jasper AI.

  • Multimodality - Marketing is a fundamentally multimodal discipline that spans text, video, images, audio, and more. Jasper needed to deliver a seamless and integrated multimodal experience to address the needs of modern marketing teams.

  • Deep verticalization - Performing marketing tasks at the highest level of quality requires depth. Jasper is building the first LLM for marketing that has a deep understanding of marketing best practices and the nuances that marketers face in their daily work.

  • Orchestrating multiple LLMs - Jasper’s large models need to work in concert with other, smaller, purpose-built LLMs for tasks like writing ad copy or generating images and must securely integrate their customer’s proprietary data. Jasper leverages proprietary LLM APIs from OpenAI and Anthropic alongside open and custom LLMs based on families like Llama 2 and Mixtral.

  • RAG - Marketing cannot be done without significant amounts of context and data. Retrieval augmented generation (RAG) across vast quantities of public and private data is an essential part of building a marketing solution that caters to the subtleties of a particular customer’s business.

  • Velocity with a lean team - AI and marketing are rapidly evolving. To stay at the forefront, Jasper AI’s team must rapidly adapt to incorporate new models, new data modalities, and new methods. Business and technology breakthroughs depend on rapid experimentation and iteration, especially in AI.

  • AI evolution and future-proofing - Jasper’s prior AI solution was not originally designed to support LLM workloads. The team needed a highly-customizable platform that was ready for the LLM era and built in an open, extensible way for future evolution.

The Solution

After an extensive evaluation to find the right partner to support the next evolution of their market-leading platform, the Jasper AI team selected Anyscale. Anyscale’s end-to-end AI Platform provided critical infrastructure capabilities that serve as a backbone to Jasper’s AI Copilot.

  • Iteration velocity - As the number of models developed and managed by the Jasper team increased, the complexity of upgrades and fixes grew to the point where even minor changes would take hours. Now, teams can upgrade models in 10 minutes, allowing for significantly more iterations.

  • Flexibility - Anyscale made it easy to deploy, customize, and scale multiple models. These models include popular LLMs like Llama 2 and Mixtral, image generation models like Stable Diffusion and ControlNet, as well as a long tail of traditional models for tasks like image recognition, optical character recognition, and beyond. Anyscale simplifies the management and orchestration of all of these models, allowing them to be used in concert to power applications.

  • Rapid fine-tuning - Fine-tuning plays an important role in improving model quality, and Anyscale’s platform provided Jasper with breakthrough capabilities to rapidly fine-tune LLMs for specific use cases. For example, all companies that advertise on Google have to follow Google ad guidelines, and those guidelines change periodically. Rather than spending weeks tuning the perfect prompt to a general-purpose foundation model, which would be slower and more expensive, the team is able to customize a smaller, highly-specialized model to do a far better job.

The Impact

The Jasper team has effectively created the “blueprint” for the future of applied AI, both in terms of what it offers to users functionally, and in terms of how it’s delivered architecturally.

Jasper didn’t bolt a chatbot onto an existing app or put in a “generate image” feature -- they took a holistic approach. They put AI everywhere across their portfolio, giving users the potential for a true “10X” transformation rather than a 10% incremental improvement.

Beyond that, they’ve taken an innovative approach to getting the most out of their LLMs. Behind the scenes, they seamlessly orchestrate multiple LLMs in their tech stack, including OpenAI, Anthropic, Mixtral, and Llama 2. As a result, users see consistent performance across different LLMs for different tasks, while minimizing complexity and cost by intelligently using the right LLMs for the right tasks.

Key benefits the Jasper AI team experienced from leveraging Anyscale:

  • 6x faster product iteration - This translates to increased productivity for ML and Ops experts and ultimately faster product delivery. “We can’t let our ops stack dictate our innovation pace. And having something that was built for LLMs makes it much easier for our internal AI and ops teams,” said Guhan Venguswamy, Head of Platform at Jasper AI.

  • Future proofing - AI technology changes rapidly. With a user base well over 100 million, and growing, Jasper AI is not in a position to re-architect their stack every time a new model or methodology emerges. Anyscale’s open, extensible architecture and extensive integrations allow Jasper to tailor the most suitable and effective models depending on users’ needs, now and adapt them in the future as their goals evolve. 

  • Multimodality for user productivity - “Marketing is fundamentally multimodal. Text generation has become a parlor trick. Modern marketers need imagery, text, audio, video, and more,” said Suhail Nimji, VP of Business Development at Jasper AI. “Anyscale provides the flexibility to leverage all data modalities across a wide range of models."

The Future of AI in Marketing is Bright

The visionary team at Jasper AI sees AI becoming even more pervasive and powerful in marketing and is committed to adding more multimodal capabilities and exploiting new models, techniques, and technologies behind the scenes to be the AI Copilot that marketers can’t live without.

Ready to try Anyscale?

Access Anyscale today to see how companies using Anyscale and Ray benefit from rapid time-to-market and faster iterations across the entire AI lifecycle.