MoCon 2023 Recap
ChatGPT's release in November 2022 sparked a revolution. Generative pre-trained transformer (GPT) models had been around for a few years, but ChatGPT enabled helpful, user-friendly access to the masses. The ease of access to GPT models from OpenAI, Anthropic, and the open-source community has led to an explosion in GPT interest and a rapidly changing landscape. In this talk, you'll get up to speed on using generative AI in your own products. We'll cover key terminology and concepts in LLMs as well as common tooling that is used for programmatically interacting with LLMs. Further, we'll see patterns for implementing generative AI in your applications along with difficulties to watch for. Whether you're curious about how to use generative AI in your own product or just wanting to build your own chatbot to console you before AGI takes over, this talk will direct you on how to understand the latest and greatest in generative AI.
Developers are implementing GenAI solutions rapidly to bring intelligent solutions to their customers. This introduces challenges in model selection, controlling costs, and data orchestration. This session evaluates the right level of AI/ML required for your solution, and using pay-as-you go services to help reduce cost. It demonstrates using serverless services to orchestrate large amounts of data to interact with a GenAI solution efficiently. After this session, you will have a practical understanding for approaching GenAI that you can take back to your team. The solution demonstrated in this session is a cost effective approach to using AI/ML, GenAI, and serverless to handle complex real world problems at scale.
Embracing the future of database technology involves redefining the way we approach data interaction. In this talk, we'll journey through the creation of two groundbreaking tools at PingCAP–TiDB Serverless and Chat2Query – and how they're designed to simplify and revolutionize developers' interaction with databases. TiDB Serverless is our response to the call for simpler database management. As a scalable, distributed SQL database built for cloud-native environments, it eliminates server management tasks, freeing developers to focus on what truly matters – their code. We’ll shed light on the journey of creating a tool that stands up to the challenges of serverless computing. Next, we delve into Chat2Query, a tool designed to make large-scale databases accessible via natural language inputs. With this, developers who aren't SQL experts can engage with databases more intuitively. We’ll discuss the fascinating process of developing this tool, breaking down language processing and query optimization challenges.
Feature Store has become an interesting topic and a newly minted cornerstone of modern machine learning infrastructure. Distributed caching, though still relatively young at about two decades old, feels like a mature establishment. This talk will show you the unmistakable ties between the two by revisiting the waves of development that led us from one to the other, all the while expanding this little “niche” called in-memory data store. What properties of the systems have remained consistently important, and what requirements have changed? What shall we expect moving forward with bigger models, more features, and more data? This little journey down history lane and beyond will give you a clue.
What makes developers tick? Join us as we gather a group of industry experts to explore practical strategies and challenges in fostering technology adoption. From open-source projects to proprietary libraries, we discuss everything from rapid experimentation to continuous feedback. Gain insights into empowering developers, while dealing with real-world business risks such as resistance to change, competing priorities, and technical debt. Together, we will discover the secret sauce behind developer-led adoption!
Among users of CYDAS PEOPLE, an HR Tech SaaS in Japan, there was a challenge in the situation where a few HR professionals were receiving inquiries about tens of thousands of HR-related procedures. We therefore expanded the HR FAQ functionality that we already offered to answer the questions we receive from employees using FAQ data and ChatGPT. I will talk not only about how LangChain and Momento were very helpful, but also how I had to be careful when using LLMs in production applications, such as stabilizing output content, protecting personal data, preventing hallucination, and countering prompt injection.
Search is hard! Yet, search underpins systems that we rely upon daily. Search is also evolving rapidly, especially with the recent advances in applying ML to search ranking. In this talk, we cover the building blocks of a flexible search system: indexes and an ML-learned ranking system to sort and combine search results for queries across the indexes. We start with the basics of indexing, followed by a deep dive into inverted and neural indexes, and introduce a "system of experts" construct to combine results from multiple indexes along with a ranking system to provide the ultimate search experience for the end users.
What if you could discover a new way of storing and searching data that is more powerful, more elegant, and more natural than anything you have seen before? What if you could learn how this new way is inspired by the sacred geometry of nature, the patterns and structures that are hidden in plain sight? What if you could realize that you have already encountered something similar in your past, something that captures the essence and meaning of words? This is what vector databases are. In this talk, we will explore what vector databases are, how they work, and why they are useful for AI applications. By the end of this talk, you will have a better understanding of vector databases, and how they can help you create more intelligent, more beautiful, and more human applications.