Skip to content
Neurasense

Research

Published research

Peer-reviewed articles and preprints from Neurasense and its research collaborators, with direct DOI and arXiv links.

  1. Artificial intelligence based personalized student feedback system 'Sisu Athwala' to enhance exam performance of medical undergraduates

    Thilanka Seneviratne, Supun Manathunga, Wathsala Idirisingha, Kosala Somaratne, Kosala Marambe, Udaya Dangahadeniya

    Peer-reviewed articlePublicationPLOS One · 20(12), e0336154Published4 December 2025InstitutionUniversity of PeradeniyaDOI10.1371/journal.pone.0336154

    A feedback system for medical undergraduates, built on retrieval augmented generation over a custom knowledge base to reduce inaccurate answers from the underlying model. It gives each student feedback on their MCQ and SAQ performance, along with guidance on stress management and study strategy. It was evaluated both by expert student mentors and by the students who used it.

    Reported study results

    15 expert student counsellors evaluated the feedback; 25 students reported on its usefulness.

    Related capability: Retrieval-augmented generation

    Read the full article

  2. Artificial intelligence assisted automated short answer question scoring tool shows high correlation with human examiner markings

    HMTW Seneviratne, SS Manathunga

    Peer-reviewed articlePublicationBMC Medical Education · 25(1), 1146Published5 August 2025InstitutionUniversity of PeradeniyaDOI10.1186/s12909-025-07718-2

    A scoring tool that uses a large language model to mark short answer questions. It extracts the key components of a student's answer, applies the scoring rubric supplied by the instructor, and writes individual feedback. It was tested on a systematic pharmacology course, where its marks correlated closely with those of human examiners.

    Reported study result

    Correlation reported between AI scores and scores from two human examiners.

    Related product: AES

    Read the full article

  3. Aligning large language models for clinical tasks

    Supun Manathunga, Isuru Hettigoda

    PreprintPublicationarXiv · arXiv:2309.02884PublishedSeptember 2023PreprintarXiv:2309.02884

    An alignment method for medical question answering, called expand-guess-refine, which combines instruction tuning with in-prompt techniques. It reached 70.63% accuracy on a subset of the USMLE dataset without additional parameters or training data.

    Reported benchmark result

    Accuracy reported on the study's selected USMLE dataset subset.

    Related capability: Clinical AI alignment

    Read the full article

  4. Retrieval augmented generation and representative vector summarization for large unstructured textual data in medical education

    Supun S Manathunga, YA Illangasekara

    PreprintPublicationarXiv · arXiv:2308.00479PublishedAugust 2023PreprintarXiv:2308.00479

    Sets out retrieval augmented generation as a way to attach a customisable knowledge base to a language model, so that answers in a medical context can be grounded in a known source rather than in the model's own memory. It also presents a combined extractive and abstractive summarisation method for large unstructured text, using representative vectors.

    Method overview

    The paper's indexing and retrieval workflow, shown in sequence.

    Related capability: Retrieval-augmented generation

    Read the full article

Authors are listed in publication order. Entries marked with a university affiliation were carried out with collaborators at that institution.

Explore related products