The application of large language models (LLMs) to specialized fields, such as Satellite Communications (SatCom), presents unique challenges due to the extensive and cutting-edge knowledge required. SatCom encompasses a wide range of technical details, protocols, and operational guidelines that must be addressed to produce effective and accurate models for practical use. This paper presents a fine-tuning approach for adapting 7-billion-parameter instructed LLMs (Llama-3v and Mistral) to SatCom, using a proprietary corpus sourced from the European Space Agency (ESA) consisting of domain-specific PDF documents. The confidential nature of this corpus imposes constraints on both model training and evaluation, demanding a sensible text extraction pipeline capable of handling complex structures, such as tables, to preserve critical information. Our fine-tuning methodology employs a carefully configured process, followed by an automatic evaluation framework using a curated Q&A set tailored to SatCom. Models were created in both non-quantified and 8-bit quantized formats, ensuring feasibility for desktop-level inference. The fine-tuned models demonstrated a 6,6% improvement over the baseline LLM, as well as significant gains when compared to retrieval-augmented generation (RAG) methods. These results indicate a promising advancement in the development of LLMs for domain-specific applications within the SatCom field.
Structured multi-agent debates among Large Language Models (LLMs) have emerged as a powerful paradigm for enhancing reasoning reliability and argumentative coherence. Motivated by the European Space Agency’s (ESA) interest in trustworthy AI for space operations, this study proposes a moderated, domain-adaptive multi-agent debate framework applied to the high-stakes domain of satellite communications (SatCom). Specifically, it assesses (i) the efficacy of structured deliberation against single-agent baselines, and (ii) the impact of model heterogeneity versus homogeneity. A single-agent baseline is compared against a multi-agent framework deploying a moderator and two domain-specialized experts. These systems utilize local 70B-parameter LLMs in homogeneous (Llama-3.3) and heterogeneous (Llama-3.3 + DeepSeek-R1 + Qwen-2.5) configurations, all augmented with a shared, curated Retrieval-Augmented Generation (RAG) corpus combining academic institutional sources and ESA material from the Nebula portal (SatNex V programme). Outputs from 213 technical queries are evaluated via LLM-as-a-judge across three phases: baseline proficiency, strategic reasoning, and executive readiness. Single-agent systems lead in encyclopedic tasks, where retrieval suffices over deliberation. However, both multi-agent configurations outperform in strategic reasoning, with heterogeneous debates achieving superior performance in executive scenarios by victory margins of up to 2.75 points on a 10-point scale. These results validate architectural diversity as a decisive factor in resolving high-complexity technical conflicts. Ultimately, this work delivers a generalizable, fully traceable deliberation framework suitable for real-world, mission-critical environments. Code, prompts, and evaluation data are publicly available at https://github.com/amozo-es/multi-agent-debate/ .