From Chatbots to Agents: Building Autonomous Workflows with Databricks Mosaic AI

Find out how Mosaic AI can be leveraged for various business applications. Whether you need an intelligent customer portal, monitoring for cyber-attacks or creating predictive maintenance models this course provides essential knowledge.

Lakehouse Monitoring provides real-time insight into data processes and model performance, alongside centralized governance via Unity Catalog and MLflow for robust GenAI or LLMOps operations.

Transforming Industry Solutions: From Chatbots to Agents

Chatbots have proven particularly useful for managing production and inventory in manufacturing by sending commands directly to the underlying system, then using machine learning (ML) techniques to detect bottlenecks and make decisions that improve efficiency and decrease costs. This simple demonstration shows how Databricks Mosaic AI allows organizations to integrate AI capabilities into practical industry solutions.

React provides a user-friendly, interactive interface for querying the bot. A FastAPI backend serves as the mediator between React and Mosaic AI agent models-serving endpoint, receiving messages and returning task-specific responses from React frontend.

The Mosaic AI Agent Framework makes it possible to develop and deploy production-grade AI agents using Python, with tools for creation, evaluation, governance, as well as unified workflows that utilize GenAI models.

Mosaic AI offers an expansive catalog of both open source and proprietary GenAI models through the Mosaic AI Model Catalog. You can easily switch large language models (LLMs) without changing application code, enable usage tracking for usage monitoring purposes and set guardrails to control model scalability – these features make GenAI accessible quickly and securely while scaling applications to deliver enterprise-grade LLM performance. In addition, Mosaic AI Gateway makes integrating, deploying and fine-tuning large GenAI models easy on enterprise data directly for business intelligence/analytics purposes.

The Mosaic AI Agent Framework: Building Autonomous Workflows

Mosaic AI offers a full-stack platform to speed the development, deployment, and monitoring of generative AI applications. This includes no-code tools for high performance modeling as well as scalable model serving, routing, observability monitoring systems for production-grade agent systems.

Mosaic Agent framework offers a fast path to deployment for compound AI systems, featuring an easy unified interface for managing, governing and evaluating models. It enables faster and more reliable MLflow pipeline while agents can be deployed without manual Python notebook generation; furthermore Mosaic Agent Evaluation reports quality metrics directly to human stakeholders while being integrated with MLflow experiment tracking/debugging functions.

AI Gateway is a centralized tool designed to administer rates, permissions and credentials across all model APIs (external or internal). This enables rapid experimentation to find the ideal model for every use case while Gateway Usage Tracking and Inference Tables record who called which models – providing chargebacks and audit trails when necessary.

The AI Gateway is a highly scalable and performant model serving service, optimized for real-time inference with low latency. Utilizing serverless compute, it provides a centralized interface for deploying, governing, and querying models across platforms – including custom models registered through Unity Catalog as well as foundation models from workspace model registries such as scikit-learn, XGBoost, PyTorch, Hugging Face transformer models.

Scaling Enterprise-Grade AI Agents

Gain production-grade AI with agents rooted in enterprise data. Agents can automatically ingest and synchronize data, execute commands, retrieve information from multiple systems, generate outputs such as reports or recommendations and quickly build and deploy agents that tether themselves directly to enterprise data ensuring reliable performance with scalable compute and storage capacity. With Mosaic AI you can rapidly build and deploy agents that connect back into enterprise information ensuring reliable performance with scalable compute and storage capacity.

Centralized Governance and LLMOps

The Agent Framework simplifies building and deploying agents by supporting any model type – including non-deep learning models such as regression, LightGBM, and XGBoost – including classical non-deep learning models like regression. Furthermore, built-in evaluation for agents uses production traces with judges and scorers used during development to measure quality on production traces with real-time monitoring provided by Lakehouse which allows you to quickly debug and optimize agent-driven applications.

Mosaic AI makes it simple and straightforward to deploy compound AI systems by offering one interactive interface where users can create, customize, deploy and update complex AI models with prompts, LLMs, tools and deployed agents.

Mosaic AI provides centralized governance and observability of GenAI operations, or LLMOps. Azure Databricks for discovery and logging; AI Gateway to manage GenAI models on any model endpoint; MLflow versioning infrastructure-as-code to implement robust LLMOps processes; as well as Unity Catalog which simplifies management by centralizing data, models, and tools into one location.

Databricks Apps: Deploying Internal Data & AI Applications

Mosaic AI Agent Framework helps developers quickly develop production-grade agent systems. With its straightforward API and easy user interface, this API supports developers in creating domain-specific, high-quality AI agent systems that meet common AI use cases such as an evaluation system aimed at checking that an agent performs well without bias or any associated costs; additionally it offers tools for deploying and querying these agents.

AI applications are rapidly transitioning from human-triggered workflows to autonomous goal-driven agents that coordinate, plan, reason and act at scale. These agentic AI systems are more powerful than static prediction models and designed for more diverse use cases.

Databricks Apps makes it simple and fast for teams to build and deploy internal data and AI apps. The platform supports popular Python libraries like Dash, Shiny, Grado, Streamlit and Flask for developing diverse data applications. In addition, development teams can leverage preferred development practices like Git version control and CI/CD pipelines while it runs on automatically provisioned serverless compute instances with built-in governance via Unity Catalog as well as fine-grained security support.

Combining Mosaic AI Vector Search, Foundation Model APIs and Databricks Workflows enables organizations to rapidly build AI models that align with enterprise business goals. IT can then deploy GenAI+RAG applications using these capabilities and reduce operational complexity and costs by dismantling siloed machine learning (ML) projects and automating manual ETL processes.