OceanAI: A Conversational Platform for Accurate, Transparent, Near-Real-Time Oceanographic Insights

TL;DR

OceanAI addresses hallucination and data-access barriers in ocean science by grounding conversational queries in real-time NOAA datasets through a function-calling, multi-agent architecture. It couples a coordinating LLM with modular data-retrieval, computation, and visualization components that access CO-OPS, CORA, and CRW datasets, producing verifiable outputs with provenance. Empirical comparisons show OceanAI outperforms baseline AI chat interfaces on structured NOAA data tasks, due to robust grounding and transparent metadata. This framework enables near-real-time, reproducible ocean insights with potential for operational decision support in hazards, ecology, and water-quality monitoring.

Abstract

Artificial intelligence is transforming the sciences, yet general conversational AI systems often generate unverified "hallucinations" undermining scientific rigor. We present OceanAI, a conversational platform that integrates the natural-language fluency of open-source large language models (LLMs) with real-time, parameterized access to authoritative oceanographic data streams hosted by the National Oceanic and Atmospheric Administration (NOAA). Each query such as "What was Boston Harbor's highest water level in 2024?" triggers real-time API calls that identify, parse, and synthesize relevant datasets into reproducible natural-language responses and data visualizations. In a blind comparison with three widely used AI chat-interface products, only OceanAI produced NOAA-sourced values with original data references; others either declined to answer or provided unsupported results. Designed for extensibility, OceanAI connects to multiple NOAA data products and variables, supporting applications in marine hazard forecasting, ecosystem assessment, and water-quality monitoring. By grounding outputs and verifiable observations, OceanAI advances transparency, reproducibility, and trust, offering a scalable framework for AI-enabled decision support within the oceans. A public demonstration is available at https://oceanai.ai4ocean.xyz.

Figures (12)

• Figure 1: Comparison of model responses to “What is the maximum water level in Boston in 2024?”. OceanAI returns the correct NOAA-verified value (2.79 m MSL) with full metadata, while GPT-4o omits the value, Gemini-2.5 Pro miscalculates, and Grok-3 declines.
• Figure 2: Function-calling workflow in large language models (LLMs). The LLM parses a user query (e.g., What is the sea level in Boston in 2021?) into a structured API call (e.g., get_ssh(2021, Boston)). The call is executed externally, and the returned results (e.g., figures or statistics) are combined into the final response openai_function_call.
• Figure 3: OceanAI system architecture. The unified agent-based pipeline executes fixed modules for web retrieval (via Brave API), document search (via Chroma), and curated NOAA media. Function calling is dynamically triggered based on query semantics, providing structured access to NOAA datasets such as SST, sea level, and CORA via parameterized back-end functions.
• Figure 4: Standardized function output schema in OceanAI, including text, images, json_data, and others (metadata: location, units, time coverage, source).
• Figure 5: OceanAI output for a sea level trend query comparing Boston and Virginia Key in 2022. Data from NOAA CO-OPS includes monthly mean sea levels, annual change rates, and plotted trends for both locations.