Project summary

• Evaluating clinical decision support AI and how safe LLMs are before they give health advice to patients

• AI researcher who studied medicine then moved to AI Engineering, postgrad student, Cairo

• 2 papers accepted NeurIPS 2026 Sydney, my first NeurIPS

• GenAI4Health workshop 12 Dec: https://neurips.cc/virtual/2026/loc/sydney/177660

• MuslML workshop 7 Dec: https://neurips.cc/virtual/2026/loc/sydney/166663

• no lab, no grant

Goals

• GenAI4Health paper: most tests give the model the full patient case in one message but real patients don't act like that. The model has to ask and be evaluted for its asking. I propose 3 new measures for this.

  MuslML paper: a model safe in one clinical area is not always safe in another. The way the models failed for emergencies wasn’t like how it failed for mental health.

• present both posters and find collaborators for doctor written triage dataset.

Funding

• flight only, 5-13 Dec around $1,500,Cairo-Sydney

• min $1,000

• NeurIPS pays already the registration and hotel

If it fails

• If I can't present in Sydney (in person only), I will release my spot so other one can use it.

Raised

• $0 cash

• BlueDot rejected, Emergent Ventures rejected and university can't support.