An AI-assisted R&D platform transforming biomedical literature into structured evidence, transparent scoring and research intelligence — from a 126-study MSc prototype to a 94,000+ paper discovery engine.
V1 established that the architecture works. The current programme expands the same logic across a much wider PubMed-derived research universe, with strict triage before expensive deep extraction.
The prototype deliberately separates the existence of a positive finding from the strength of the underlying evidence.
The V1 dataset contains enough human and preclinical evidence to expose the translational gap rather than flattening all studies into one category.
The platform asks a more defensible question: what kind of evidence exists, how strong is it, how confidently was it extracted, and where does preclinical promise fail to translate into robust human evidence? The goal is an inspectable evidence layer that researchers can interrogate — not an automated medical verdict.
Programmatic literature retrieval and peptide-hit mapping from biomedical sources, preserving source identifiers and metadata.
Local LLM extraction converts unstructured abstracts into a defined evidence schema covering study type, design, sample size, outcomes, statistics and safety.
Deterministic normalisation, evidence scoring and confidence scoring create a comparable analytical layer across heterogeneous research.