I build and evaluate AI systems for evidence-based medicine. At JournalClub.net, a US health-education and research platform based in Princeton, NJ, I lead a team of three engineers. We build Clinical Pearl, a production retrieval-augmented generation (RAG) platform that answers clinical questions for more than 1,000 clinicians, and I am currently building AI Appraisal, which helps readers critically appraise medical research articles.
My research interest is trustworthy language models for clinical use: methods to evaluate, ground and calibrate LLMs and RAG systems so their outputs can be checked against evidence. I co-developed CAMEO-AI, a framework for appraising AI randomized controlled trials, and PEARL-G, both presented at EBM Live, Oxford (2026). My MS thesis at FAST-NUCES combined vision transformers and language models for radiology report generation on MIMIC-CXR. I have three peer-reviewed publications in applied NLP and multimodal learning.
I am looking for a fully funded PhD position (Fall 2027) in trustworthy and clinical AI. Please get in touch if my work fits your group.
Building AI Appraisal for JournalClub.net to help clinicians appraise medical research articles.
Launched Nabz, an Urdu-first AI health-triage companion, built for the Alibaba Cloud AI Hackathon (Healthcare track), and Daulat, an evidence-based PSX investing companion.
Presented CAMEO-AI and PEARL-G at EBM Live, Oxford.
Paper on sentiment-aware image captioning published in Computers, Materials & Continua.
Promoted to Lead AI Engineer.
EEBERT published in IEEE Access.
Urdu emoji-aware sentiment paper published in IEEE Access.
Completed my MS in Data Science at FAST-NUCES, Islamabad.
Joined JournalClub.net as ML/AI Engineer on Clinical Pearl.
Received the IEEE eLearning Course Trial Award (IEEE Xplore Challenge for Researchers in Pakistan).
K. R. Narejo, H. Zan, D. Oralbekova, K. P. Dharmani, O. Mamyrbayev, K. Mukhsina
IEEE Access, 2024.
Conference posters & talks
CAMEO-AI: a 190-item appraisal framework for AI randomized clinical trials, with a bibliometric and quality review of 2,826 AI/ML RCTs (2020–2025) using RoB 2, CONSORT-AI and SPIRIT-AI. Poster, EBM Live, Oxford, June 2026 (co-author).
PEARL-G: evaluation framework for clinical AI. EBM Live, Oxford, June 2026.
Urdu-first AI health-triage companion · Alibaba Cloud AI Hackathon 2026
A health assistant for Pakistani families. It runs Urdu voice triage one question at a time, reads lab reports and prescriptions from photos and explains them in Urdu, and keeps a per-patient Medical Vault with family profiles and a health timeline. Emergency red flags stop the conversation immediately, and a printable English handoff summary is ready for the doctor.
React · Vite · FastAPI · Qwen (qwen-plus, qwen-vl-plus) · SQLAlchemy · installable web app
A calm, long-term investing companion for the Pakistan Stock Exchange. Users plan a goal with bear, base and optimistic paths (inflation, dividends and step-ups included), analyse any stock with a fair-value range and plain-language pros and cons, track their real portfolio (cost basis, P/L, XIRR), and ask questions answered from their own holdings. Every number comes from a tested engine and carries its source; missing data is shown as unavailable, never guessed.
Python · valuation & portfolio engine · LLM Q&A grounded in user data · automated market-data refresh
Production RAG platform for clinical decision support used by 1,000+ clinicians. I lead the team; we cut query latency by 40%, improved answer relevance by 20%+ and reduced deployment failures by 35%.
AI-assisted critical appraisal of medical research articles on JournalClub.net, helping residents and clinicians judge study quality and bias.
LLMs · evidence-based medicine · appraisal checklists
Disease–drug knowledge base
Evidence-backed knowledge source for Clinical Pearl that links diseases and drugs using FDA drug data, RxNorm and SNOMED CT, used to ground drug and treatment answers.
Appraisal framework and interactive evidence site for AI randomized clinical trials: 2,826 trials reviewed for risk of bias and reporting completeness.
Evidence synthesis · RoB 2 · CONSORT-AI · SPIRIT-AI
Visual abstracts
Automated pipeline that turns clinical guidelines and papers into structured visual summaries (risk rail, decision flow, key points), verified across 75 urgent-care topics.
AI point-of-care clinical engine for urgent care (Society for Academic Urgent Care Medicine): structured consults, AI search, ambient history-taking, visual diagnosis and clinical decision rules.
Agentic system that tests IT access controls over the full evidence population, isolates exceptions with an evidence trail and routes low-confidence cases to a human, with an evaluation harness.
End-to-end pipeline from transactional streams to an Azure SQL database, with a churn model and Streamlit dashboard.
Azure Data Factory · SQL · Streamlit
Experience
Lead AI Engineer, JournalClub.net (Princeton, NJ; remote) Lead a team of three on Clinical Pearl and AI Appraisal; evaluation and observability, hiring and mentoring; HIPAA-regulated data.
ML / AI Engineer, JournalClub.net — Clinical Pearl Fine-tuned LLMs (GPT, LLaMA, Mistral) for clinical-trial applications; built RAG and multimodal RAG systems.
Machine Learning Engineer, Cplus Soft, Islamabad Generative AI for question answering and summarisation; LangChain workflows.
Machine Learning Trainer, AI Lounge / NUST, Gilgit-Baltistan Designed and taught a data science and AI curriculum.
NLP Teaching Assistant, FAST-NUCES, Islamabad
Education
MS in Data Science, FAST-NUCES, Islamabad Thesis on radiology report generation with vision transformers and language models.
BE in Computer Systems Engineering, Mehran University of Engineering & Technology, Jamshoro CGPA 3.72 / 4.00.