mongodb-developer/GenAI-Showcase
GenAI Showcase
该项目是一系列生成式 AI 实现，专注于 AI 代理、检索增强生成（RAG）流水线和向量搜索集成的开发。它提供了一个将托管云数据库连接到语言模型的框架，以创建上下文感知应用。
该项目涵盖了使用多步推理和外部工具完成任务的自主代理的编排。它包括使用高维嵌入进行语义检索的实现，以及使用模型无关的提示词（prompting）以确保不同大语言模型之间输出的一致性。
其他功能包括使用地面真值（ground-truth）评估框架来衡量 AI 性能的准确性和可靠性。该项目还演示了用于存储和管理应用数据及向量嵌入的云数据库账号设置。
Features
- Retrieval-Augmented Generation - Connects MongoDB vector search with language models to provide grounded and context-aware responses.
- Agentic Tool Orchestration - Orchestrates autonomous agents that call external functions and tools to complete multi-step tasks.
- AI Agent Development - Provides tools and environments for developing autonomous AI agents capable of multi-step reasoning.
- RAG Pipelines - Combines external data stores with language models to improve response accuracy via RAG pipelines.
- Semantic Vector Search - Implements semantic retrieval using high-dimensional embeddings to find documents based on conceptual meaning.
- Autonomous AI Agents - Provides frameworks for building autonomous agents that use reasoning and tool-use to complete complex tasks.
- Vector Database Integrations - Integrates vector databases to store and manage document embeddings as a source of truth for AI context.
- Ground-Truth Scoring - Implements methods to compare model outputs against ground-truth datasets using scoring metrics.
- AI Evaluation Frameworks - Uses AI evaluation frameworks to measure the accuracy and reliability of model outputs.
- Prompt Templates - Uses standardized templates to ensure consistent output across different large language models.
- RAG Context Retrieval - Retrieves relevant document segments from a vector database to ground language model responses.
- AI Observability and Evaluation - Measures the accuracy and reliability of generative AI outputs using standardized evaluation frameworks.
