Hi everyone and welcome to my session. It is called large language models for better developer learning of your product. My name is Babur. I'm a Devop advocate at company called Atacama. Also I am ex Microsoft. I have been working in Devrel over five years. I'm also active open source contributor to Apache Software foundation. If you would like to have a chat or connect me on LinkedIn Skord and make a connection, I would be more than happy to have an online session to answer your question and address about Devrel and AI in Devrel how we are leveraging AI in Devrel and I have been also speaking in many events last year and this year in 2024 last year I've been delivered sort of different in person and online sessions at various conferences. You can see the pictures depicted from some previous sessions. So today's agenda is about how as we know that developers we are highly interested in AI technologies. I assume that more than 60% of the developers globally already working on or learning about AI and generative AI. In this session we'll discuss how the developer relations meets generative AI this year and how these changes AI changes effecting on dev rail some of the concentration you need to take into consideration when you are integrating AI tools into your Devrel strategy. I will also walk you through these tools and we will try to address the main question is the AI wave currently going on is a threat or an opportunity for Devrel teams nowadays growing so the bunch of records such as legal papers, academic studies, news, technical guides and books can be automated using AI. Large language models can be applied, as we can see, to various use cases and industries. For example, OpenAI is GPT four, a powerful LLM used for a wide range of NLP tasks and chart GPTO. We know that set record the fastest growing user base in January 2023 prove that large language models are here to stay with us longer. As I am also a developer and I have been working with Python a long time, I decided also to build my own applications using new AI technologies. What about quickly summarizing your content? Get the information you need in real time from private large unstructured documents in your Dropbox. For example, the same tool can be used with OneDrive or Google Drive. I decided to make my job easier when I'm creating invoices or summarizing my content using my own private tool that connects a Dropbox and analyze my documents. Or next. This is another app to find real time discounts sales prices from various online markets around the world. I connected to real time Amazon APIs to fetch some discounts, deals, coupons, information and then it makes me easier to find out these discounts I am interested in. And you can also advance this feature by adding some alerts. When there is some discounts you can get some discount information from the AI application. With another application you can think about LLM app that provides real time alerts for significant document change or updates. Let's say you are working with marketing campaigns and this system can monitor various aspects such as content chains, campaign performance metrics, or the audience engagement. Real time alerts enables marketing teams to respond quickly to changes to make sure that these campaigns remain on track. Or the last one I would like to demo one of the applications is effortlessly extracting and organizing unstructured data from PDF's, Docs or other unsustainable information more into SQL tables in real time. This example, as you can see, extracts data from unstructured files and stores into PostgreSQL table. Also it transforms a user query into SQL query which is then executed on postgreSQL tables. So here as you can see, we are progressing quite fast based on the four applications I was able to build and it takes me for each building application like one or 2 hours and it's already ready to use. Let me know in the comments if you find this application useful, I will be more than happy to provide the source code for them and then you can also give a try yourself. So let's switch our attention back to the developer relations as we call Schwarz dev rail. Right, developer relations, exactly what it means as a marketing policy that plutoizes some relationship with developers. And for those who doesn't know in depth, what is a dev? What are the Devor publicists do? For example, Devrel can act as a bridge between your company and its technical audience. Our primary goal is to build a strong, engaged community around the product or any developer technology. We provide education, support and we try to foster the engagement and we try to simplify the developers learning, experience and challenges they face. And some people ask like when exactly Devil can be helpful for the company? Let's assume that if you're building a developer oriented product, that's where Devwell can help. Assume that marketing team cannot reach the right audience when they are demanding highly technical content and they actively avoid the usual sales and marketing channels. Because the rails or the robot is a technical person, they are more like expertise in providing some examples and sending messages for your audience. Or another thing is maybe product managers are struggling to understand the new industry trends or without being expert in this domain. Or sometimes engineering teams are super busy in building a product and maybe they might not have a time or skills set to do everything that Devrel does. So. And from that perspective, our devrel divides into four pillars. Mainly like we do developer marketing. We try to understand who we are targeting, what kind of developers for a product, and we make sure that they have information and tools to make the decision. Also developer enablement, developer advocacy and the community. Also our part of responsibility and as you can see developer relations after it introduced, it enabled also marketing community and other things. For example, we are creating and maintaining always the process where our developers can have a common goal relations. Developer relations enable developer education and foster developer experience and support and developer success. As you can see for the developer education, we create sometimes documentation, tutorials, video, videos and guides for developer experience. We improve always API design and SDK experience. Also we probably get some feedback always while they are using our SDKs. And we also support the developer success once we know what the dev rail is. And now let's bring your attention and how AI changes developer relations nowadays, AI I think will be the accelerator of pre existing developer relations trends. For example, AI assistant documentation for production use cases. For example, AI can act as a copilot for us, taking our routine and boilerplate tasks. In this context, AI can help speed up the documentation process by inspecting our APIs and the code. Also, it's helping us nowadays creating code samples and also supporting in real time some developer requests. You don't have to answer yourself. An AI chatbot can answer these questions. It's also in the support context. For example, a well implemented bot always can handle simple support requests. Or or if you have an AI chatbot developers, instead of going through the enormous documents docs files, they can also search in the search bar specific information I have started using chat GPT for example, generate short descriptions of new articles, new variation of titles from my blog post, even some article outlines why I still have to guide and fact check the AI machine because it is saving at least a few hours of my work. But at least I am one who is fixing and advancing the AI solutions. Imagine being able to summarize community posts, discord channels or slack channels. I can almost guarantee we will begin seeing soon like community copilots that can help to coordinate between different channels and outlets. However, as you can see, generative AI is creative, but not as creative as we are humans. It also doesn't do well with personal experience or realistic examples, especially initiate topics like ours data processing pipelines in Python. While you can try to push to do so. It doesn't always understand real life struggles with replicating human experiences and it doesn't do well with extending piece of context. It always struggling to go deeper topics so you will often find repeating the same level of definition of your topic while you are asking questions. Here's a table of things how the AI is helping me nowadays. Let's say for example last week I pushed the sample report to GitHub showing how to build real time data processing pipelines in Python. This also involved with some of the number one use cases just above since I was using chartGpt while writing the code because the simple scenarios nowadays GPT can create somewhat close to reality. The most powerful thing I think the AI you can feel feed some context documentation information to your AI and provide some searching functionality, let's say on your doc page. This is like how we can approach in Devrel to help the developers to find easily the information. We already seen some work being done with support bots or askdocs functionality for our device documentation or the API descriptions. Let's have a look at building AI chatbot that can help the people understand your developer documentation more easily and answers user questions. They can find answers quickly without needing human intervention, right? Which might be speeding up the workflows and improves overall developer satisfaction with the pathway team. What we did, we integrated this ask me bot into discord server where our pathway community they are. You can ask questions about your specific information to find out easily the documentation details or code samples. This definitely allows developers to specific question from a prompt. It saves time from parsing the pages of the documentation or contacting developer relations represented directly from the codewise of simple if you navigate our repository open source repository pathway, you can find many examples. As I demonstrated at the beginning, one of the examples also looks quite simple, how to connect the AI chatbot to our documentation. It's very easy. As you can see we are just connecting to docs data by using built in connectors and we have some libraries to make it easy to calculate vector embeddings by chunking the big amount of data into chunks and feed this data into to the discord servers in real time. And let's say it's called also differently rug approach retrieval argument that the approach as you can see in the diagram, it highlights the common architecture for it. While you are ingesting some data from APIs or files databases, what you do, you just build a data pipeline that processes data, transforms it and calculate some embeddings. As we know vector embeddings and also after calculating vector embedding, it stores for fast retrieval to the vector storage like a vector database. And then you can start to build your application on the top of it. That provides some search bar with the backend service that accepts user queries with questions. And then it does again like from the query calculates vector embeddings and from the vector embeddings it finds relative clause vectors we stored in the previous step in vector databases. That's how the common rag approach works. But if you are trying to build your application or feed the context, custom context to AI application, not easy enough, you need to know nowadays a lot of technologies and frameworks, sometimes it's, well, confusing. And here we can see a lot of technologies nowadays building around AI applications such as frameworks, APIs, foil and so on. And while with a team of pathway when I was working in the past, we tried to build our own also the framework to help the developers to reduce this job. I mean you don't have to know all the technologies and tools to build applications. Our LNM app provided by Pathway, it's fully open source where you can replace all these technologies and knowledge by using a single application. Some of the simple application we introduced our develop experiences by using the pathway open source technology. For example, it means we you can also reduce go to market time. It's a lower cost because open source is for free and it says highly security like a context where you can also run it on the top of your custom llms without using public LLM provider. And of course it's using also under the hood pathway technology where you can connect to any real time data sources. Nowadays as open source, we are supporting open following sources like it can structure data, structured semi structured live data. You can ingest data from your Docs page, you can ingest your data from PowerPoint, any PDF's or slack channels. To analyze your developer experience better, here's a list of key features LLM application offers. For example, it can index real time documents without using any vector storage or vector databases in real time. It means it also reduces infrastructure overhead. From the architectural perspective, it simplifies much nowadays. Use the emerging technologies with lnms. As you can see, simple architecture we are demonstrating everything is managed by the single framework where you don't have to know how these things internally works with a shorter lines of code. You are already building some developer experiences for this discord server. For example, you can connect to your internal local files or external maybe GitHub markdown files to ingest the data and we can connect to multiple providers LLM providers publicly available like OpenAI, somewhere from Facebook or Google. So other stuff is fully managed like vector indexing, chunking the information and feeding this data into the discord servers. There are still some challenges to make these LLM applications or AI chat post to the production level. It's easy to provide some examples. We are still testing our skill server applications. The issues we have seen like besides of natural languages. Sometimes we are facing hallucinations and there are also constant latency. For example, you never know when you can get answer from OpenAI because they don't have SLA's. There's no average time responding right. You can expect the real response from the AI OpenAI on time. Other things problems is offline evaluation. Of course, when you're writing for example unit tests or testing in the documentation or code samples for correctness, it's impossible to evaluate it without connecting to public openaPI servers. And sometimes it's open. The LLM providers, the LLMs, they respond differently at each request and it's impossible to test and make sure that everything is working fine. As you have seen, it's easy to make something cool with lnms, but it's very hard to make something production ready with them. So if you interested in observing our open source framework in Python for building applications, come with square code. It will bring you to our source code. I have shown you shown already to you and you can try to run discord simple application. Maybe it might help you to run your own discord chat AI bot response based on the documentation you have. Here's the takeaways from my sessions. As you can see, integrating the AI into software development and Devrel is better as it's accelerating our existing productivity rather than it's completely disrupting us and AI representing both significant opportunities and challenges for Devrel. While it can automate also tasks like technical documentation support, that leads to increased efficiency and potentially it reaches developers experience. And I think also AI will enable new class of users like creators who can tap into productivity that previously only available to software engineers. Non software engineers also can do some engineering works by using low code and no code trends and start also think about AI integrations for your developer product product. Thank you for attending my session. If you would like to have questions, feel free to ask me on LinkedIn or leave your question in the chat. I will be more than happy to address them. Thanks. Take care. Bye.