Project summary

I am building an AI tool that reads patient records with names and other identifying details removed. It looks for possible links and turns them into research hypotheses: ideas that researchers can test.

Useful clues can be scattered across a patient's notes, laboratory results and medication history. The tool would bring those clues together, suggest a research question and show the records behind the idea. It would also explain what is missing or uncertain.

The aim is to help researchers find useful questions they might otherwise miss. This project focuses on generating those questions and helping researchers decide which deserve further study. Finding out whether a proposed link is true would require separate research.

What are this project's goals? How will you achieve them?

With full funding, we would build and evaluate the tool over 18 months. We would start with suitable public medical datasets that have had identifying details removed and that we have permission to use.

Months 1–6: build an early working version and try it on one dataset. Each suggested question would show the records that led to it.

Months 7–12: improve the questions, reduce repeated or unsupported ideas, and check published research to see what is already known.

Months 13–18: evaluate the tool on up to three suitable datasets from different areas of medicine, depending on data availability and permissions.

Each hypothesis would describe the patient group, the possible link and why it may be worth studying. It would also show missing information and other possible explanations.

Independent medical reviewers would assess the AI's questions alongside questions that researchers produce from the same records. They would judge whether the questions are clear, useful, supported by the records and possible to investigate. We would also measure how long it takes to produce and review useful questions.

We would deliver a working tool, a list of research questions ranked by their value for further study, and a report on how well the tool performed. We would share our methods and evaluation results within the datasets' sharing rules. I would report progress and adjust later work based on what the first stage shows.

How will this funding be used?

I am requesting up to $300,000 for the full 18-month project. A minimum of $30,000 would let us begin with a six-month pilot on one suitable public dataset.

The full budget is:

$144,000 for software and data engineering. This is enough for the equivalent of two engineers working for 18 months at an estimated $4,000 per person per month.

$54,000 for clinical research leadership: $3,000 per month for 18 months to guide the medical questions, study design and response to reviewers' feedback.

$36,000 for independent medical reviewers to assess the research questions.

$30,000 for access to AI models and computing resources.

$20,000 to prepare and check the data used to build and evaluate the tool.

$16,000 for reports, project coordination and administration.

These estimates total $300,000. Hiring and spending would begin after funding.

For the smaller $30,000 pilot, we would use $18,000 for research and engineering, $5,000 for independent medical review, $3,000 for AI and computing, $3,000 for data preparation, and $1,000 for documentation and administration. The outputs would be an early working tool, research questions linked to their supporting records, and an evaluation report on one dataset.

Funding between $30,000 and $300,000 would support more datasets, more independent review and more development time, as the early results justify. All stages would focus on generating and evaluating medical research hypotheses.

Who is on your team? What's your track record on similar projects?

My name is Abdullah Hamdan. I am a medical doctor and the founder of Curans. Curans helps medical students practice clinical thinking and communication before treating real patients. It now has more than 35,000 learners. Building it taught me how to turn a clinical need into a tool that people use.

I would lead the medical side of this research project. Enad Abu Zaid supports technical development. We would recruit independent medical reviewers and additional engineering support as funding allows.

This research project is at an early stage. In exploratory work using public Cancer Imaging Archive data, we noticed a possible link between recorded PD-L1 status and brain lesion size. This is an early clue that can lead to a research question. It needs further review and independent testing before we draw medical conclusions.

What are the most likely causes and outcomes if this project fails?

The AI may misread records, make weak assumptions or suggest ideas that are already well known. A medication or laboratory result can provide a clue without proving that a patient has a particular disease. The same pattern can also have several explanations.

We would show the source behind each idea and separate what the records say from what the AI suspects. Medical reviewers would check the proposed links. We would compare the questions with published research and with questions produced by researchers using the same information.

The project could fail if its questions are too broad, unreliable or time-consuming to review. We would report those findings openly and use them to decide whether further work is justified.

How much money have you raised in the last 12 months, and from where?

We have received no funding in the past 12 months ($0). There are no funding sources to report.