EVE: A Domain-Specific LLM Framework for Earth Intelligence

2026-04-16 · Source: cs.AI updates on arXiv.org · Field: Science & Research — Environmental Science & Earth Systems, Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

Earth Virtual Expert (EVE) is an open-source, end-to-end framework for developing and deploying domain-specialized Large Language Models (LLMs) for Earth Intelligence (EI). The core component, EVE-Instruct, is a 24B parameter model built on Mistral Small 3.2, optimized for reasoning and question answering in Earth Observation (EO) and Earth Sciences. EVE includes curated training corpora (2.8B open-access tokens, 10.7B synthetic instruction tokens) and the first systematic domain-specific evaluation benchmarks, covering multiple-choice QA, open-ended QA, and factuality. The system integrates Retrieval-Augmented Generation (RAG) and a hallucination-detection pipeline into a production system, deployed via API and GUI, and has supported 350 pilot users. All models, datasets, and code are openly released on Hugging Face and GitHub, demonstrating strong performance on domain-specific tasks while preserving general capabilities.

Key takeaway

For AI Engineers developing domain-specific LLMs, EVE demonstrates that a targeted approach combining domain adaptation, curated data, and robust evaluation can yield superior performance without relying on larger models. You should consider adopting a similar end-to-end framework, including synthetic data generation and a RAG pipeline with hallucination detection, to build reliable and efficient specialized AI systems for complex scientific domains.

Key insights

EVE provides an open, end-to-end framework for domain-specialized LLMs in Earth Intelligence, outperforming general models.

Principles

• Domain adaptation improves performance without increasing model size.
• Interleaving instruction and long-form text preserves general capabilities.
• RAG and hallucination detection enhance factual reliability.

Method

EVE fine-tuned Mistral Small 3.2 using a mixed data strategy, combining general-domain replay with synthetic EO/Earth Sciences text, then applied Online Direct Preference Optimization for alignment.

In practice

• Use a two-pass chunking strategy for RAG documents.
• Employ LLM-as-a-judge for open-ended QA evaluation.
• Implement rolling summarization for conversational memory.

Topics

• EVE Framework
• Earth Intelligence
• Domain-Specific LLMs
• Earth Observation
• Retrieval-Augmented Generation

Code references

• ekzhu/datasketch
• microsoft/presidio
• tatsu-lab/alpaca_eval
• Kludex/uvicorn

Best for: AI Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Your AI Doesn’t Know Anything. And That’s Not the Model’s Fault. — AI hallucinations stem from infrastructure limitations, not model intelligence, necessitating robust knowledge layers like graphs.
• TerraMARS: A Domain-Adapted Small-Language-Model Pipeline for Mars Terraforming Literature — Domain-adapted Small Language Models can extract structured knowledge from scientific literature for complex planetary science applications.
• Connect the Dots: Training LLMs for Long-Lifecycle Agents with Cross-Domain Generalization Via Reinforcement Learning — LLMs can achieve continuous self-improvement and cross-domain generalization by learning to self-update context via end-to-end RL.
• HPC-LLM: Practical Domain Adaptation and Retrieval-Augmented Generation for HPC Support — Domain adaptation and RAG enable smaller LLMs to match larger models in specialized HPC support with fewer resources.
• TADI: Tool-Augmented Drilling Intelligence via Agentic LLM Orchestration over Heterogeneous Wellsite Data — Domain-specialized tools and structured prompts drive analytical quality in agentic LLM systems for technical operations.
• Domain-Specific Data Synthesis for LLMs via Minimal Sufficient Representation Learning — DOMINO inductively synthesizes domain-specific data for LLMs by learning minimal representations from reference examples, bypassing explicit domain descriptions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.