Summary
This guide provides detailed recommendations for selecting AI models suitable for Aid4Mail filtering, classification, and review workflows. It outlines various models, including their strengths, weaknesses, and appropriate use cases. The document begins by presenting the fastest practical recommendations, highlighting models like Mistral Small 3.2 24B for high-volume binary review and Qwen 3.6 27B Dense for accuracy-first multilingual tasks. It explains key metrics such as F1, Automation Yield, and INCONCLUSIVE rates, which are crucial for evaluating model performance. The guide also discusses the decision-making process between using cloud, enterprise cloud, or offline models, emphasizing the importance of data governance and operational efficiency. Additionally, it provides a shortlist of practical models based on deployment type and specific needs, ensuring users can make informed choices based on their requirements. The guide concludes with considerations for privacy, reproducibility, and language support, making it a comprehensive resource for Aid4Mail practitioners.
