February 2024 Summaries

deepset Cloud is a scalable, accessible, and user-friendly option for implementing applications with Large Language Models (LLMs) inside, providing a flexible and unified environment for integrating various LLM components into a single platform, enabling rapid prototyping, visibility, and transparency. It offers standardized yet customizable components, helping to mitigate the steep learning curve of Gen AI, while also addressing the cost in time and engineering resources associated with leveraging LLMs in a full DIY manner. By streamlining the way AI teams build real-world applications with LLMs, deepset Cloud is compressing time by offering an indispensable toolkit and infrastructure that augment internal platforms and toolchains, bringing useful templates and trusted reference implementations based on deepset's years of experience in implementing scalable solutions for enterprise customers.

The Haystack Enterprise Platform, previously known as deepset AI Platform, is tailored for organizations integrating large language models (LLMs) into their products by offering a balanced approach between a universal black box solution and a customizable toolchain. It provides a unified development environment for AI teams, facilitating rapid prototyping and experimentation with standardized and customizable components, along with monitoring and analytics tools to align development with business objectives. Designed to overcome the steep learning curve and resource costs associated with LLM implementation, deepset offers flexible pricing and professional AI services to help organizations optimize costs and launch pilot applications efficiently. By leveraging deepset's extensive experience, the platform streamlines the creation of LLM-driven applications, providing templates and reference implementations to mitigate the risks involved in designing and implementing an LLM-based architecture.

In the context of earnings calls, AI is being utilized to analyze the impact of artificial intelligence on various companies, particularly those in tech and financial services. A Retrieval Augmented Generation (RAG) system was developed to extract insights from live earnings calls, leveraging OpenAI's Whisper model for transcription and a large language model (LLM) such as GPT-3.5 for generating conversational answers. The RAG system involves indexing pipeline steps including data collection, transcription, embedding, and indexing, followed by a query pipeline that uses both vector and semantic retrieval to select relevant documents and prompt the LLM with user questions. This approach allows for the generation of condensed answers grounded in recent data, overcoming training data limitations. The development of this RAG system was made possible using deepset Cloud's AI platform, which streamlines the development life cycle by providing a unified environment and customizable components.