Air operations rely on complex rules, established procedures, and time-critical analysis under limited connectivity and strict security constraints. In such environments, analysts must combine written doctrine with images, often without access to external computing resources. This paper studies offline large language models as decision support tools, deployed in isolated and restricted environments to give analysts access to doctrinal knowledge that remains traceable to its original sources through natural language interaction. We describe a modular retrieval-augmented architecture suitable for operation without Internet connectivity, supporting both text and image input from technical manuals. As a first step toward evaluating this architecture, we report a pilot study with four image analysts of the Brazilian Air Force, combining (i) a doctrinal knowledge assessment based on their electronic-target identification doctrine, comparing human and proposed system performance on the same test, and (ii) a measurement of the cognitive workload involved in manually producing a reconnaissance target report (Relatório de Missão de Reconhecimento - REMIR) without AI assistance. The results show a demanding manual task, especially in terms of mental demand (6.0/7) and effort (5.0/7), while the proposed system matches the human score (8/10) and completes the assessment in 7.1 minutes (compared to a human average of 26.5 minutes), establishing a baseline for future AI-assisted evaluation. Finally, we describe a future evaluation protocol to systematically compare manual and AI-assisted workflows.
Figures & tables
Figure 1: Architecture for offline multimodal language model deployment in air operations decision support.
Figure 2: User interface mockup illustrating a decision support session where the analyst submits a technical query and receives a structured response grounded in the MCA 200-5 doctrine, complete with source citations.
Participant
Role
Experience in image analysis
Familiarity with MCA 200-5
Analyst 01
Officer
>7 years
High
Analyst 02
Officer
4 – 7 years
Moderate
Analyst 03
Enlisted
>7 years
Very high
Analyst 04
Enlisted
4 – 7 years
Moderate
Table 1: Profile of the image analysts participating in the pilot study.
Participant
Score (correct / 10)
Time (min)
Analyst 01
10
26.0
Analyst 02
10
42.0
Analyst 03
10
23.0
Analyst 04
8
15.0
Proposed System
8
7.1
Table 2: Knowledge test results: Human analysts vs. Proposed System.
Model & Regime
Score (correct / 10)
Avg Time (s)
Total Time (min)
Standard Regime (with Justification)
qwen2.5:14b
8
121.91
20.3
deepseek-r1:14b
8
227.90
38.0
llama3.1:8b
5
44.28
7.4
Optimized Regime (Answer Only)
qwen2.5:14b
8
42.86
7.1
Table 3: Local model evaluation under different prompting regimes.
Figure 3: Inference acceleration (speedup factor) achieved by the optimized prompting regime relative to the standard prompting regime across the three local models.
Workload Dimension
Mean ± SD
Mental Demand
6.0 ± 1.2
Physical Demand
4.0 ± 1.8
Temporal Demand
4.3 ± 1.5
Performance
2.0 ± 0.8
Effort
5.0 ± 2.2
Frustration
4.0 ± 1.4
Table 4: Subjective workload of the manual REMIR task (Raw NASA-TLX, N=4 , scale 1–7; SD = Standard Deviation).