Callimacus Redefines Websites and E-Commerce with Millisecond Inference Powered by Groq

When Italian luxury fashion group Brunello Cucinelli began exploring artificial intelligence back in 2021, they weren’t chasing trends. They were looking for a way to fundamentally rethink how customers experience the digital landscape, and how integrating AI could augment—rather than replace—human creativity. That early spark grew into Solomei AI, an independent research venture focused on applying AI across industrial processes and digital experiences. The company is led by Francesco Bottigliero, a long-time Brunello Cucinelli executive who now serves as its CEO.

By 2023, Solomei AI’s team, including an interdisciplinary group of engineers, mathematicians, philosophers, artists, and creative technologists, identified an opportunity that reshaped their roadmap. “We had the idea to innovate the way websites are designed and built,” Francesco explained. The result was Solomei AI’s first project: Callimacus.

Named for the ancient Greek poet, scholar, and librarian, Callimacus is an AI-driven platform designed to redefine the way websites are developed and managed, offering a new way to think about how e-commerce sites are created and run. “As the core of our technology, Callimacus replaces static web pages with dynamic, interactive experiences that are tailored to each visitor,” Francesco says. “It's not just about displaying content; it's about creating a personalized journey where every interaction is unique and purposeful to the user.”

Rethinking the digital shopping experience

Solomei AI set out to make the digital shopping experience as responsive and intuitive as an in-store boutique associate. “Traditional websites, as well as e-commerce sites, still act like static catalogs,” Francesco says. “Everyone sees the same content, they get irrelevant recommendations, and their real-time intent is ignored. If someone walks into a boutique asking for a blue blazer, no one would lead them to eyewear—but digital commerce does that every day.”

After launching the first Callimacus-powered website, Solomei AI received overwhelmingly positive feedback from peers in the fashion industry. Interest quickly grew beyond the company, with competitors and adjacent industries asking how they could adopt the technology themselves.

Scaling Callimacus from an internal experiment to an enterprise-ready platform, however, pushed the limits of their existing infrastructure. Delivering fully personalized, dynamic experiences required a far more powerful technical foundation.

Overcoming inference limits with Groq

Early prototypes of Callimacus relied on a mix of commercial LLM APIs and open-source models. While sufficient for experimentation, they were too slow and costly for production. Scaling Callimacus across high-traffic e-commerce experiences required:

• Low-latency inference to keep shoppers engaged: Product pages, recommendations, and assistant responses need to render instantly to avoid drop-offs and abandoned sessions.
• Stable, predictable performance across all agent layers: Ensuring consistent load times during traffic spikes, flash sales, and seasonal peaks.
• Cost-efficient compute that supports large-scale personalization: Enabling dynamic recommendations, real-time intent detection, and tailored content for millions of shoppers without breaking margins.
• Inference-optimized hardware: Maximizing throughput for high-volume catalog queries, product matching, and conversational commerce without the overhead of training-focused systems.

To bridge the inference gap, the team turned to Groq.

Francesco recognized the architectural advantage: Groq’s LPU uses a compute-in-memory design, reducing the bottleneck between memory and processing. “For training LLMs, there are options,” Francesco said. “But to run an inference-based AI business, you need something different. Groq was the right choice.”

Inside the architecture: A multi-agent AI orchestra

Solomei AI re-architected their infrastructure and began migrating their AI workloads to Groq. This included redesigning how agents communicate, routing inference-heavy tasks to Groq’s LPUs, and adjusting their pipeline to take advantage of Groq’s low-latency execution model.

Today, Callimacus uses a multi-agent architecture to turn static web pages into adaptive, real-time experiences. Each agent plays a specific role, interpreting intent, shaping tone, maintaining context, optimizing structure, and assembling the interface dynamically based on what the shopper is doing in that moment.

Supporting these agents is a modern knowledge base, including:

• A vector database for semantic retrieval
• A content graph that understands relationships between products, stories, visuals, and metadata
• A real-time event capture layer that replaces the need for tools like Google Analytics, because the system inherently sees every interaction

Instead of relying on personal data or predefined segments, the system responds to live signals—what a user views, revisits, or skips—and adjusts content, layout, and recommendations on the fly. Privacy-preserving agents ensure interactions remain secure while still delivering deeply personalized journeys.

Real-time personalized experiences at human speed

Transitioning to Groq provided a stable, efficient foundation for multi-agent inference, enabling real-time personalization at the scale needed for high-traffic e-commerce. “Groq’s LPU architecture, with on-chip memory that’s purpose-built for inference, stood out as a clear technology breakthrough,” Francesco said. “We chose Groq because delivering seamless, personalized experiences requires moving from user intent signals to content pairing in just milliseconds. Groq makes that possible.”

Callimacus now generates interfaces instantly, eliminating load times and keeping shoppers engaged from the first interaction. Consistent, low-latency token generation boosts the accuracy and quality of agentic reasoning, enabling sharper, more relevant recommendations in real time. And Groq’s price performance makes large-scale, hyper-personalized shopping journeys feasible across the entire catalog and customer base.

The result is an experience that feels fluid, intuitive, and uniquely tailored to each visitor, every time they engage.

The platform now delivers interactive guidance that feels as intuitive and responsive as a human sales associate, elevating the overall shopping experience.

And, because Callimacus understands every interaction across the experience, internal teams can use it by simply asking questions like, “How are banner views performing in Japan this week?” or “What should we change?” Then follow with, “Make the update,” and the system adjusts itself automatically.

Defining a new standard for e-commerce

What began as an internal research experiment has now become an industry-defining platform. Solomei AI launched a fully AI-driven retail platform for Brunello Cucinelli, transforming its online marketplace into a personalized shopping experience that mirrors the brand’s luxury, boutique-level attention and care. If a shopper says, “I need a dress for a gala,” the platform instantly curates complete outfit suggestions, dresses, accessories, and styling tips included. It learns each customer’s preferences over time, remembering their style and providing increasingly personalized recommendations with every visit.

At the core, Solomei’s mission remains the same: to use AI to elevate human creativity, not replace it. With Groq powering the AI layer, that vision comes to life, fast. And because Groq delivers consistent, low-latency performance across traffic spikes, regions, and workloads, Solomei AI can scale these experiences globally without ever sacrificing speed.