Multi-Agent College Admission Process System: A Review and Proposed Framework

Admission offices in Indian colleges handle a large volume of repetitive enquiries about eligibility, fees, documents and Centralized Admission Process (CAP) rounds, and conversational artificial intelligence (AI) is increasingly proposed to absorb this load. This paper reviews fifteen works covering educational chatbots, university and admission assistants, retrieval-augmented generation (RAG), hallucination and abstention in large language models (LLMs), speech recognition, human-AI interaction guidelines and data-protection law. The review finds that educational chatbots are well studied for learning and FAQ support, and that retrieval reduces hallucination in generated answers. However, admission-facing systems rarely control which knowledge is allowed into answers, rarely decline when a candidate-specific record is missing, seldom model consent for personalised outreach, and do not distinguish drafted or simulated actions from real communication. Based on these gaps, the paper proposes a human-in-the-loop framework with four mechanisms: an approval-gated knowledge lifecycle, verified-record grounding with explicit abstention for CAP queries, a consent gate before any personalised message, and an explicit action-state model that labels drafts, previews and blocked actions. The framework is presented as a design proposal together with evaluation criteria for grounding, abstention, consent enforcement and record completeness, to guide future pilot studies in admission offices.