ransform Adobe Experience Manager with AI Agents: Play, Build, Retrieve with Algolia

Transcript

I’m Debanshi. I’m an Alliance Director at Algolia, one of the sponsors for today’s event. And here we have Sajid, our Senior Integrations Director, who has built a lot of our Adobe integrations. We’re so happy to be here today. And in this session, we will showcase how enhanced retrieval augmented generation, or RAG, empowers agents with real-time, federated, context-rich answers, and how Algolia plays a role here in building these agent experiences with our retrieval layer, and building these experiences for our customers on the Adobe Stack. I need my clicker. So just introduce ourselves. Who is Algolia? Algolia is a search and discovery platform that has been around for around 14 years. We have over 18,000 customers, and have been the leader of the Magic Quadrant for the past two years in both vision and execution. We are known for our ability to support the largest enterprises with reliable performance search, while taking out the need for large search engineering teams and infrastructure, and instead empowering teams to focus on the user experience around discovery and around driving the KPIs for their business.

We have been an Adobe partner for a long time, and have over 1,600 mutual customers, using Algolia to extend Adobe Commerce and Adobe Experience Manager with advanced search, and we’ve seen this across every vertical.

This past year, we were honored to be named Adobe’s Content Supply Chain Partner of the Year, and we’re excited to announce that we were just named a Platinum Partner in the Adobe Tech Program. Over the years, to support our mutual customers, we’ve invested heavily in deeper integrations between Algolia and the Adobe Stack to support contextualized search that unifies data across all of your Adobe systems. So this includes integrations into Adobe Commerce, both the traditional integration and the app builder integration with our partner Vymo, a very deep integration into AEM with drag-and-drop components for AEM authors and indexers, and integrations into AEP that allows us to understand the context of a user, feed that into the way results are ranked. For example, here Daniel has the sneakers that are available in size 10, just for him, and also being able to take all the activity that happens in search and discovery and feed that back into AEP to then go into downstream events. For example, an Instagram campaign based on what he searched for here. As the world has shifted over the past couple years, and agentic experiences are coming to the forefront, our customers have realized that in the new world, search and discovery is actually about AI-driven retrieval, and this is the evolution that Algolia has taken with them. Getting the right information for both customers and agents has never been more important and important to get right, meaning the same foundational principles that we’ve had for search and discovery now apply to agentic retrieval. So first, speed. We don’t expect agents to make up the answer. They need access to the freshest enterprise data from enterprise systems, and they need to be able to pull that right data very quickly. When we start thinking about agent-to-agent interactions, the number of search queries explodes, and it becomes more important than ever to have the fastest search engine on the market, answering millions of complex semantic search queries across billions of records in milliseconds.

So that’s speed, right? The next one is relevance. So while the agent is focused on pulling what is most meaningful for the customer using the LLM and the model that it comes with, the retrieval layer has to make sure that the responses are also the most relevant for the business, based on all the context that the enterprise has about their customer, both as individuals and in aggregate.

So this could mean when pulling a piece of content as a part of the retrieval layer, you are actually pushing the content that is trending across your customer base at a given moment, or an order based on the five past orders that that customer has purchased.

Lastly, it’s very easy for brands to feel a loss of control when now coming on board with new agentic experiences, because the LLM is making a lot of decisions. This is where it’s very important to maintain a level of business influence in those results.

So an example of that could be making sure that you have inventory and availability of your products, and that that is being reflected in the results. Or if you’re promoting a specific brand or a specific set of products, that is also incorporated into the results. So combining what that user or that agent was looking for with what you actually want to promote as a business. We’ve seen many times, especially with people being very excited about shopping on LLMs, that you might purchase something only to receive an email 15 minutes later saying, sorry, that product was actually not in stock. And that erodes brand trust, so it’s really important to get all of this right.

So what Algolia has done to bring that speed, relevance, and control to a brand’s website, their mobile app for search and discovery, is now also being used to power generative experiences and agentic experiences, especially on your site as you are starting to create your own brand agent.

Taking this one step deeper, Algolia has built a full set of solutions to support bringing our foundational retrieval capabilities into the agentic era and to our solutions across the Adobe Stack.

Along with our search platform, we have our own agent studio to help customers build agents, and we also have the MCP tools to support their use of any other agent framework, including Adobe’s. In fact, we were the first search solution to bring our MCP server to market, and we also use this to power our Algolia Assist product, which allows users to configure search, relevance, and manage everything internally through an agentic experience. So with Algolia, you have the ability to combine any external LLM with Algolia, pulling in any other data source, such as Adobe Experience Manager content, fragments, pages, Adobe Commerce products, into your conversational experience.

So what this infrastructure enables is to us to power agents that are able to retrieve the most meaningful information for both the customer and the business. So you might be asking about a four-day trip. You need to make sure you’re receiving results in milliseconds that are relevant to both the agent or the end customer, but also to what the business has in terms of the context about that customer. That is business-influenced, KPI-driven, federated, for example, bringing both resort information and restaurant information into the answer, and contextualized, how big is this family, what is their price point, where are they traveling from, incorporating all of that into that conversational experience.

So with that, I’ll have Sajid come on and share a demo to help bring this to life. And as he does that, he will start by kind of just sharing what are the steps to get there.

Thanks, Debonji.

Hi, my name is Sajid. I run the integrations team at Algolia. My team has built a lot of Adobe and Algolia integrations, such as AEM, AEP.

I will be doing a demo on Algolia Agent Studio, where you can manage your agents for your digital experiences, powered by Algolia, Adobe Commerce, Adobe Experience Manager, and Adobe Edge Delivery for delivering your application itself. But before we begin, let’s talk about the five stages that made the demo possible.

So first thing, we need to be able to get the data out of your source systems into Algolia. So for instance, AEM, you want to get your content, pages, digital assets, content fragments into Algolia. And for Adobe Commerce would be your products and categories. Once your content is in Algolia, then the configuration of your search for your index happens, where you define your re-ranking formulas, your searchable attributes, so that you can actually get the relevant content at a basic level. Once the configuration of your search is done, then we can incorporate personalization. So not only are you trying to provide relevant content to your users, but try to find relevant content that is meaningful to me. So you can use AEP to create segments and then publish that to Algolia, so you can actually have control over some of the results for that audience.

And then next, after you have that experience set up where you have the data, you’ve configured it, personalization is enabled, you can test your search experience, because not only does a customer engage with search, you will have your AI agent also do the same. So once you have tested all of that, then you can actually build your agent and then provide the instructions to help drive the agent experience for your customer. And then the last is obviously adding the code so that your agent can then be presented in your front end to be engaged with. So let’s go to the demo. All right. Give me a second. Okay. All right. Here we go. I’m going to start off with a demo of what we’ve implemented, and then I’ll go into all of the steps to get me to this demo. So I have this grocery store demo where it’s delivered through Adobe’s Edge delivery. I have my global search as a block, and I’ve also set up universal editors to kind of allow the customers to select the index that they want to pull the data from. The global search is set up as a federated experience, so it allows me to see recipes for inspiration. And then once I select a recipe, what products do I need to buy so that I can actually make that for either lunch or dinner? I also have query suggestions as well of what other people have searched for that I can use within my search. Now, if the customer is not able to find what they’re looking for by just querying, they can then use the agent, which is also a block, Edge delivery block, that they can also engage with. Let me go ahead and clear this out. When the customer engages with the agent, they’re presented with a chat experience and a nice welcome message. Here I can actually type in, let’s see, recipes for dinner. And here the agent will take my query and use Algolia search to make some requests to the recipes index itself. Let’s actually look for Italian recipes instead. Okay, so I’m going to make another query, and here I see the agent’s provided me two recipe options. Let’s actually go with spaghetti bolognese. I want to try to see what products or what ingredients do I need to make this recipe.

So the agent here is going to go back, take a look at the products index that’s been indexed in Algolia, and here it’s actually made several requests to Algolia, multiple options per ingredient for me to choose from. So I can actually add this to the cart. But I can see that this, the recipe, is calling for me to make my pasta sauce from scratch. I may not have time for that, so I may ask the agent to provide me recommendations on pasta sauce. And the reason why I’m doing this is to kind of talk about the control that DiBanchi mentioned earlier, where a business may be running a campaign around Prego, and they want to surface that as one of the pasta sauce in position one. So here you can see that Prego is showing up as the number one position. So now let’s talk about how all of this is put together.

So we’ll start off with the source system. And here we have AEM, where my recipe content fragments are actually managed.

My team has built the AEM Algolia connector. Now, Algolia is an API-first company, so we have API clients, so you can build your own custom connector. But we have one that we built through the integrations team, which allows a customer to easily set up, install, and configure the indexer. Right here, I have my indexer for the grocery stores. And here I can click on Edit. This is where I can enter in my credentials for Algolia and the properties I care to index into Algolia. Now, we do provide additional advanced features as additional…

enrichment that you can do as well on your data before you send that to Algolia. You take this configuration, you map it to the folder of where all your content exists, and then your publish action or unpublished action that’s available through AEM. We listen to that, so on a publish, we’ll add the record, and on an unpublished, we’ll remove the record. We also provide a full indexing console, so if you already have published content and you want to index those pieces of content without having to go to each content and publish them, you can use this console to do that itself. So now, once my content has been… my editorial content has been indexed into Algolia, then I can use the Algolia commerce integration as well to index my products and categories. Now, my source system content and product data is now in Algolia, and this is where the second step is where you kind of configure your index itself. So I have my products. This is where I can take a look at my ranking formulas to see if this is what I want to use to find the content or the products that I’m looking for, searchable attributes. I may have a lot of attributes that I index into Algolia, so I may want to narrow down the number of attributes that I search through. After you’ve configured just, let’s say, this basic level of configuration settings, you can test out your experience when you’re searching for, let’s say, pasta to see if you’re actually getting relevant responses. Now, this is keyword. We also have vector search as well, so if you want to turn on semantic search, you can, and you can actually see how well certain products came up, whether it was keyword, 100%, or vector, or whatever that mix was. And same thing goes for the products as well. Oh, actually, sorry, recipes as well. Now, going back to the Prego example, where I wanted to surface my business rule to promote Prego, I have a rule that says that if I’m looking for pasta sauce, I want to position the Prego product at number one position. So I’ve done that, and this is kind of where I’ve informed my agent to respect that promotion. So that’s step two. Now, step three, which was personalization, I tried to do it, but I just ran out of time, so I wasn’t able to achieve step three in my demo. So that could be done through our agent as well through client-side tools.

So now let’s go and build the agent. So you go through the, sorry, in the Algolia dashboard, you go to generative AI, and here you can see Agent Studio. We do have MCP server, as DaVinci mentioned, so if you have your own agent platform that you want to use, you can actually use our MCP servers to expose those indices so your agent can use them. But in our demo, we’re just going to use the Algolia agent.

Now, when you go into the agents, you can create a new agent, or you can actually modify or manage your existing one. So since I already have a grocery shopping agent, I’ll go ahead and go into that. And when you’re here, you can actually define your instructions. So what is this agent? What is the role of the agent? What are some of the guidelines it needs to follow? And then also, what are the outputs as well? And I have my tools already set up with Algolia search, and I have two indexes or indices that I have configured, and a brief instruction for the agents to understand when to use those indices.

I also have a provider, so we’re your bring your own model. So I’ve set up OpenAI as my provider, and I can select the model that I want to use for that. And once that is configured, then you can kind of start testing your agent and see if it’s giving you the right responses so you can see if you need to make any adjustments to the instructions. So recipes. And you can see that when the request is made, the agent has used the tool Algolia search. You can take a look at the arguments, what was the query, and the number of hits, which I have actually defined over here, number of results.

And if you say, let’s see, get products, Flop will wrap. You can see that it’s going to make multiple requests for each of the ingredients and return at least five items per request. And once you’ve configured and you’re happy with the responses or the configuration of your agent, then the next step is to publish it, and the very last step is actually integrating it into your front end. Now, we do provide an agent API URL if you already have a chat framework and you just need to use that agent, you can. You can use our agent API URL, or you can use our accelerator code, whether you’re using React instance search or instance search JS or even Vercel SDK. You can copy it, modify it for what you need, and with that, you should be ready to go. So this is just one of many types of experiences you can create using Algolia as the retrieval layer. We have customers exploring this with travel and helping customers find the right cruise, vacation, auto and fitment use cases, fashion. The possibilities are endless, as you can imagine. We would love to chat with you more. We have a QR code a couple slides down. Thank you. So feel free to take a picture of this if you’d like to schedule time with us. We also have a small booth right there, so we’re happy to chat with you after the session.