The demand for mental health services far outstrips the available resources. Could AI chatbots help bridge this gap? Researchers are exploring this possibility with innovative systems like SuDoSys, a large language model (LLM) chatbot designed to offer structured psychological counseling. Unlike generic chatbots, SuDoSys follows the World Health Organization's Problem Management Plus (PM+) guidelines, a structured seven-step framework for addressing mental health challenges. This framework allows the chatbot to navigate conversations with a clear direction, offering stage-aware support through the counseling process. Traditional LLM approaches to mental health counseling often involve directly fine-tuning models with conversation data. This can create chatbots that, while capable of generating responses, might lack the structured approach needed for effective counseling. SuDoSys addresses this limitation by integrating the PM+ guidelines, ensuring conversations progress logically and address key aspects of the client's concerns. It uses a combination of an LLM, a stage-aware instruction generator, a response unpacker, a topic database, and a stage controller. This sophisticated structure allows the chatbot to understand the client's current stage in the counseling process, access and update relevant information, and generate coherent, empathetic responses. To rigorously evaluate SuDoSys, researchers developed a novel automatic evaluation technique. They used transcripts from actual PM+ counseling sessions and, with the help of another LLM (GLM-4), simulated client interactions with SuDoSys. These simulated dialogues were then evaluated for coherence, professionalism, empathy, and authenticity using GPT-4. Both objective evaluations using GPT-4 and subjective assessments by human participants suggest that SuDoSys performs comparably to existing fine-tuned models. It offers an advantage in conversation coherence, ensuring the chatbot stays on track and provides targeted support. Importantly, SuDoSys achieves these results without the need for extensive fine-tuning on specific counseling data. This makes the system more cost-effective and potentially easier to scale and adapt. While promising, challenges remain. Improving the system's understanding of complex human-computer interactions and further refining its performance with real-world counseling data are key next steps. SuDoSys offers a glimpse into a future where AI might play a more significant role in providing accessible and structured mental health support, though much work remains to be done to ensure effectiveness and safety.