H2O.ai recently announced the launch of its new multi-agent platform h2oGPTe, which combines generative and predictive AI models to provide businesses with more consistent responses. Sri Ambati, the founder and CEO of H2O.ai, stated that one major demand from businesses for AI agents is to maintain consistency in responses.
The h2oGPTe platform utilizes H2O.ai's proprietary models, Mississippi and Danube, while also providing access to a variety of other language models. The platform supports operation in various environments, including air-gapped, on-premises, and cloud systems. Ambati emphasized that by combining generative and predictive AI, businesses can confidently use these agents without compromising security.
Key features of h2oGPTe include:
1. Multimodal Document AI: Provides accurate query responses from secure internal data sources (such as document libraries and knowledge bases). Supports generating structured JSON output, suitable for contract summarization and compliance data extraction.
2. Audio and Visual Analysis: Extracts structured data from audio files, images, and handwritten documents, enriching the visual domain. Audio models support multilingual transcription and translation, while visual models can perform content validation.
3. Coding Assistant: Helps developers quickly generate foundational code for projects, supporting common programming languages - providing code completion and documentation generation, speeding up the development process from concept to prototype.
4. Autonomous Intelligent Agents: Agents can automatically execute multi-step workflows, such as web research and data modeling. They can generate multi-page PDFs based on real-time data, ensuring transparency.
5. Citation Verification and Transparency: Utilizes advanced retrieval-augmented generation (RAG) technology to support embedded document citations, enhancing the transparency and verifiability of AI responses.
6. Customizable Security Barriers: Provides fine-grained access management and response limitations to ensure data security and compliance in sensitive environments.
7. Intelligent Model Routing: Dynamically selects the most suitable model based on real-time assessments to optimize efficiency and performance.
8. Model Risk Management: Transparent model evaluations combined with human feedback and automated testing to enhance compliance and explainability.
Ambati pointed out that generative AI excels in content generation, particularly in code generation, while predictive models provide greater assistance in scenario simulation. Predictive models ensure more consistent responses from agents by learning patterns from data. He noted that while interactions among humans may vary, there is still an expectation for consistent replies, which highlights the value of combining predictive AI with generative AI.
The h2oGPTe platform is particularly suited for businesses in finance, telecommunications, healthcare, and government that need to handle multi-step tasks. The platform's agents can process multimodal data, such as charts, and answer questions like “Should my company sell more dolls this year?” by considering the company’s historical financial data and market trend information.
Additionally, H2O.ai's agents can generate PDF documents that include charts and tables to visually present information and ensure data source citations for traceability. The platform also features built-in model testing capabilities, including automatically generating questions and simulating different inquiries to test the consistency of agent responses. Users can view which databases, models, or workflows the agents have utilized through a dashboard.
In the ongoing wave of AI agents, ensuring the value provided by these agents, including consistency, reliability, and accuracy, is crucial. H2O.ai's approach of combining generative and predictive models is an exploration, and other companies are also seeking ways to ensure AI agents do not disrupt businesses, such as the Agent Graph System launched by xpander.ai and the Agentforce Testing Center released by Salesforce.
Highlights:
🌟 The new platform h2oGPTe launched by H2O.ai combines generative and predictive AI models to provide more consistent responses.
📊 The platform is especially suitable for businesses in finance, telecommunications, healthcare, and government that need to handle complex tasks.
🛡️ H2O.ai focuses on data traceability and agent testing to ensure the reliability and accuracy of the agents.
