Writing
Ideas, experiments and lessons from building AI systems — by Tasawar Abbas Khan and the team. Published from Pakistan, read globally.
Why your medical AI needs to explain itself
A model that is right 90% of the time but cannot say why is unusable in a clinic. What I learned building explainability into diagnostic systems — and why accuracy alone never earns a doctor's trust.
Why your medical AI needs to explain itself
A model that's right 90% of the time but can't say why is unusable in a clinic. Why accuracy alone never earns a clinician's trust.
Grounding an LLM in five million papers
Inside MedReason-RAG: how live PubMed retrieval and FAISS embeddings stop a model inventing citations.
Grad-CAM in production: making a vision model show its work
Heatmaps are easy in a notebook and hard in a deployed app. Notes from shipping Grad-CAM inside DermaAI.
Surrogate models: turning a deep network into if-then rules
The idea behind my IEEE paper: pairing a DNN with an XGBoost surrogate so predictions arrive as readable rules.
Building a health assistant that works in Urdu
Most medical AI assumes an English-speaking user. What it took to make DermaAI genuinely bilingual.
From notebook to Hugging Face Space
Taking a trained model out of Jupyter and putting it somewhere people can actually click it.