Keeping up with health and pharmaceutical research is an ongoing challenge for university faculty. New studies, clinical trial results, regulatory decisions, and industry announcements appear across different publications every day. Identifying which developments deserve attention takes time, particularly when their relevance and supporting evidence vary considerably.

The AI Learning Centre (AILC) has developed an automated research intelligence workflow for the University of Nicosia’s School of Life and Health Sciences and Pharmacy programme

The Power of an Agentic Platform

The system uses an Agentic Platform to coordinate source monitoring, content preparation, AI evaluation, and delivery. Two specialised AI stages handle different parts of the analysis. The first assesses the available evidence and academic relevance. The second writes concise summaries of the selected stories.

Automated filtering and similarity analysis connect these stages, reducing repetitive coverage and keeping the final briefing manageable. The workflow is configured to run every two days, reviewing articles published within the previous 48 hours.

The Use Case: Making Research Monitoring More Useful

A laboratory finding, a clinical trial result, and a regulatory approval can all attract similar headlines. For an academic audience, however, understanding the stage and strength of the evidence is essential.

The objective was to create a briefing that explains what happened, how well the development is supported, and why it may matter for faculty research, teaching, grant awareness, or collaboration.

The workflow’s research interests include AI in pharmacy education, drug discovery, natural products, precision nutrition, healthy ageing, peptide therapeutics, herbal medicine safety, and pharmaceutical regulation.

Workflow Process Overview

1. Source Monitoring and Content Preparation

The process begins by retrieving a list of RSS sources from a structured data table. The source list can be maintained separately, allowing publications to be added or removed without changing the workflow’s processing steps.

Recent entries are collected, and their titles, links, authors, publication dates, and available feed content are standardised. Formatting is removed, and records without usable text are excluded before AI evaluation.

The original links remain attached throughout the process so readers can return to the source material.

2. AI Evaluation and Evidence Classification

The first AI agent evaluates each article using five weighted criteria, producing a total score out of 25:

1. Scientific Validity, up to 7 points: The rigour and quality of the evidence described.
2. Research and Faculty Relevance, up to 6 points: Its usefulness for research, teaching, scholarly activity, and academic priorities.
3. Translational and Regulatory Impact, up to 5 points: Its implications for clinical research, pharmaceutical development, regulation, or public health.
4. Evidence Traceability, up to 4 points: How clearly the reporting identifies its underlying sources.
5. Novelty and Strategic Signal, up to 3 points: Whether it introduces a new finding or a development worth monitoring.

The scoring instructions separate the importance of a topic from the strength of its evidence. An article about a major health issue may be highly relevant while still describing preliminary research.

Each development also receives an evidence classification: Clinical, Translational, Preclinical, or Regulatory. A separate Weak category identifies unusable or insufficiently supported content.

Articles must score at least 15 out of 25 to proceed.

3. Grouping Related Coverage

The same research announcement may be reported by several publications. To reduce repetition, the platform uses AI embeddings to compare article titles and group stories with similar meanings.

Within each group, the highest scoring article is retained as the representative story. The selection is then ranked by score, using coverage frequency to break ties, and up to ten stories move to the summarisation stage.

This helps keep the briefing focused on distinct developments.

4. Writing for an Academic Audience

The second AI agent prepares a structured entry for each selected story. It produces a short introduction, a concise summary, a key takeaway, and an explanation of the development’s relevance to faculty research or teaching.

The agent also assigns a topic and matches the story to specific research interests where there is a clear connection. These labels help readers identify material related to their own areas of work.

The writing instructions call for neutral language, caution around preliminary findings, and clear distinctions between research evidence, regulatory updates, and industry commentary. A further validity check is designed to exclude entries that lack enough substantive information to summarise responsibly.

5. Briefing Delivery and Website Publication

The selected entries are assembled into a formatted Health & Pharma Intelligence briefing. Each story includes its score, evidence classification, topic, summary, key takeaway, research relevance, and original article link.

The briefing also provides a priority reading overview, highlighted research themes, and a guide explaining the scoring criteria. An estimated reading time saving offers an indication of the difference between reviewing the summaries and reading the source articles.

The completed briefing is distributed through Microsoft Outlook. The same HTML output is also used to update a dedicated report webpage.

The End Result

The workflow brings collection, evaluation, and summarisation into a repeatable research monitoring process. Faculty receive a focused selection of developments with enough context to decide which stories merit closer attention.

By presenting academic relevance alongside the reported evidence, the briefing supports research awareness and helps readers direct their time towards material connected to their work. Links to the original articles provide a route to further reading and examination.

Click here to view the output