Research Areas

Each area is described by the question it addresses, the approach taken, and the contribution it is intended to make. Areas still marked as drafts are suggestions awaiting confirmation by the lab.

Reliability of AI-assisted literature reviews

Research question: How accurate are AI-assisted literature reviews — in citation accuracy, unsupported claims, and coverage — compared with manual reviews?
Approach: Inovexa Evidence organizes open-access papers into an evidence table with source references, so each claim can be traced and checked against its origin.
Intended contribution: An honest account of where AI-assisted review matches, falls short of, or exceeds manual review, published with the evidence table and code.

Reproducibility in scientific data analysis

Research question: Can another researcher reproduce the results, and how often does generated code need correction before it runs correctly?
Approach: Inovexa Reproduce helps prepare documented analysis notebooks from public datasets, with versioned data and environment specifications.
Intended contribution: Practical evidence on reproducibility rates and common failure points, with reusable notebook templates for small research groups.

Evaluating scientific reasoning in language models

Research question: How do language models perform on publicly available scientific questions — in accuracy, calibrated uncertainty, and recurring reasoning errors?
Approach: Inovexa Eval records model answers, researcher assessments, and error categories for each question.
Intended contribution: A categorized record of model reasoning failures and successes, with the assessment protocol and data published for others to inspect.