PrescQIPP

Velora: Moorfields Eye Hospital's latest AI assistant (2025)

Moorfields Eye Hospital 

Project summary

Moorfields Eye Hospital’s intranet serves as the central hub for clinical guidelines, patient records forms, research papers, and administrative policies. Given its breadth, staff spent significant time hunting for content, affecting clinical workflows, training, and ultimately patient care. To address these challenges, we designed and deployed a specialised language model–powered search assistant, optimising relevance, response speed, and user experience.

Objectives

  1. Improve Retrieval Accuracy: Leverage contextual language understanding to surface the most pertinent documents.
  2. Reduce Search Time: Shorten average search sessions by at least 50%.
  3. Boost Adoption & Satisfaction: Provide intuitive query handling, including natural-language questions and follow-up queries.
  4. Deliver Measurable ROI: Achieve cost savings via increased clinician productivity and reduced IT overhead.

This study evaluated the performance of the Moorfields Eye Hospital intranet versus a language model, called Velora. Velora only had access to published hospital guidelines that are currently available on the intranet. The evaluation focused on the efficiency, accuracy, and usability of both systems when addressing routine information queries encountered by hospital staff.

A set of ten standardised questions were developed in consultation with senior clinicians, ophthalmic trainees, and administrative personnel at Moorfields Eye Hospital. These questions were selected to reflect a representative spectrum of both clinical and non-clinical information needs within the institution. Each of the ten question was submitted to both systems independently by each user. Searches were performed in a controlled setting with no prior indication of content location. For the intranet, assessors used standard browsing and keyword search functionality; for Velora, queries were entered in plain-text format mimicking natural language input.

Initial results show that Velora can find the relevant information in a fraction of the time over the current search engine. Some pieces of information were not found using the current model whereas accessor using Velora were able to retrieve accurate information 100% of the time. The return on investment will manifest as clinical time saved per clinician as well as accurate implementation of guidelines and homogenised care. This will improve patient safety and should reduce the need for more follow up clinic appointments.