Building Digital Esther: An AI Fashion Assistant for Esther Perbandt

The fashion industry is built around identity.

For a designer like Esther Perbandt, that identity extends far beyond clothing. The brand has a distinctive visual language, philosophy, product universe, and way of communicating with its audience.

The challenge was to translate that identity into a digital experience that could interact with people naturally.

That became Digital Esther — an AI-powered fashion assistant developed with Jenna AI to help users explore the Esther Perbandt brand through conversation.

I worked on the technical implementation behind the experience, connecting conversational AI with the brand's knowledge, products, automation workflows, and WhatsApp.

The result was an AI experience that could do more than simply answer questions. It could understand context, respond to follow-up questions, help users discover products, and act as a conversational layer between the customer and the brand.

The Challenge

Traditional fashion websites are primarily built around navigation.

A customer chooses a category, opens a product page, reads the description, and decides whether to continue browsing.

But fashion discovery isn't always linear.

Someone might ask:

"I'm looking for something for an evening event."

Then:

"What would you pair it with?"

And finally:

"Show me something similar."

A useful AI assistant needs to understand that these questions are part of one conversation, rather than three unrelated requests.

For Digital Esther, the goal was therefore not simply to connect an LLM to a chatbot.

The system needed to combine:

• Brand-specific knowledge

• Product information

• Conversational context

• Product discovery

• Scope and security controls

• Automated workflows

• WhatsApp communication

• A consistent brand experience

The underlying challenge was making all of these components work together reliably.

From Chatbot to Digital Brand Experience

One of the most important decisions was treating Digital Esther as more than a generic AI chatbot.

The assistant needed to operate within the world of Esther Perbandt.

That meant the AI needed access to relevant brand and product information while also understanding what types of conversations belonged within its role.

Instead of allowing the model to answer everything, the workflow introduced layers that helped determine:

What is the customer asking?

Is the request relevant to the brand?

Does the assistant have the information required to answer it?

Should the conversation trigger a product search or another workflow?

This created a more controlled architecture around the AI model.

Connecting AI With the Brand's Knowledge

A conversational model is only useful if it has access to the right information.

For Digital Esther, I worked with a knowledge-based approach so that the assistant could respond using information relevant to the Esther Perbandt brand rather than relying solely on the model's general knowledge.

This is particularly important in fashion.

Product names, materials, collections, styling information and brand-specific terminology need to come from reliable sources.

The AI therefore became one layer of a larger system:

User → Conversation → AI → Knowledge / Product Data → Response

This approach also makes the experience easier to maintain because the underlying knowledge can evolve without having to hard-code every possible conversation.

Bringing Digital Esther to WhatsApp

One of the most interesting parts of the project was extending the experience beyond a traditional website chatbot.

I integrated Digital Esther with WhatsApp, allowing users to interact with the assistant through a channel they already use every day.

The implementation involved the Meta WhatsApp Cloud API, WhatsApp Business Account configuration, webhooks, authentication and automated message handling.

The workflow was orchestrated using n8n.

At a high level, an incoming WhatsApp message would move through a series of steps:

WhatsApp → Webhook → Message Processing → AI / Classification → Knowledge & Product Logic → Response → WhatsApp

This allowed the conversational experience to be connected to the rest of the system rather than operating as an isolated chatbot.

Designing for Real Conversations

One of the biggest differences between a demo and a real AI product is conversation.

A demo can answer:

"Show me women's jackets."

A real customer might continue:

"What about something more relaxed?"

Then:

"Show me similar ones."

Then:

"Would that work with the pants you showed me earlier?"

The system needs to understand what "that", "similar", and "earlier" refer to.

This made conversation handling an important part of the implementation.

I worked on the workflow logic around message history, conversation context and product requests so that the assistant could maintain continuity between messages.

This was also one of the areas where testing became particularly important.

AI workflows can appear to work perfectly when tested with isolated questions, but small changes to context handling can create unexpected behaviour several messages later.

Building Guardrails Around the AI

Another important part of the project was making sure the assistant stayed within its intended role.

A fashion assistant should be able to help someone discover a product or understand the brand.

It shouldn't suddenly start discussing unrelated topics, reveal internal instructions, or expose information about the underlying system.

I therefore implemented additional classification and security layers around the AI workflow.

The system could identify potentially problematic requests and distinguish between relevant brand conversations and off-domain requests.

This created a second layer of control around the language model.

The principle was simple:

The AI should be flexible in conversation, but predictable in its boundaries.

Testing Beyond the Happy Path

Building the workflow was only one part of the project.

Testing became particularly important because conversational systems don't behave like traditional forms or APIs.

A single change in one part of the workflow can affect the conversation several messages later.

I tested different scenarios including:

• Product discovery

• Follow-up questions

• Switching between product categories

• Similar-product requests

• Off-domain questions

• Security-related prompts

• Conversation continuity

• WhatsApp message handling

• Multiple users interacting with the system

I also simulated a 50-user concurrent scenario to test the workflow under event conditions.

This helped expose problems that wouldn't necessarily appear during a simple one-to-one test.

The Berlin Fashion Week Activation

Digital Esther was also part of a broader digital activation around Berlin Fashion Week.

The experience connected the physical event with a conversational digital layer.

Rather than treating WhatsApp as simply another communication channel, it became part of the experience itself.

Guests could interact with the system through WhatsApp, while the underlying automation handled the communication and routing.

This demonstrated one of the things I find most interesting about AI projects:

The technology becomes much more valuable when it is connected to a real-world experience.

What I Learned

Working on Digital Esther reinforced something I've encountered repeatedly when building AI systems:

The difficult part isn't connecting an LLM to an application.

The difficult part is everything around it.

Context management.

Data quality.

System boundaries.

Error handling.

Security.

Product logic.

Conversation design.

Testing.

And making sure all of these pieces continue working when a real person starts interacting with the system in an unpredictable way.

The project also showed me how important it is to think across both the technical and user experience layers.

My background in sales and customer-facing technology is particularly useful here. Before writing the workflow logic, I naturally think about what the customer is actually trying to accomplish.

That perspective helps turn AI from a technical feature into something people can actually use.

The Bigger Picture

Digital Esther is an example of where I see AI-powered digital experiences heading.

The future isn't necessarily about putting a chatbot on every website.

It's about creating interfaces that understand context and meet users where they already are.

For a fashion brand, that might mean helping someone discover a collection through conversation.

For another business, it could mean answering customer questions, guiding users through a service, qualifying a lead, or connecting multiple systems behind the scenes.

The technology will continue to evolve.

But the fundamental challenge remains the same:

Build AI systems that are useful, grounded in real information, connected to real workflows, and designed around the people using them.

Digital Esther was one step in that direction.