Langroid Multi-Agent Architecture for Linguistic Forensics

Design a Langroid multi-agent architecture for linguistic forensics. Define at least three agent roles (e.g., Semantic Analyzer, Style Critic, Synthesizer), their expertise, and how they would communicate via A2A protocol. How would Gemini 2.5 Pro Deep Think be specifically invoked for complex cases, and how would OpenAI o3 handle simpler ones? Detail the adaptive thinking budget mechanism.