Muhammad Tayyab F.

Key Skills

Software

Labelbox

Label Studio

Google Cloud Vertex AI

Top Subject Matter

Healthcare & Life Sciences ,Healthtech / Digital Health

Medicine / Clinical Medicine, General Practice / Primary Care

RLHF (Reinforcement Learning from Human Feedback),Medical LLM Evaluation

Top Data Types

Image

Text

Video

Top Task Types

RLHF

Red Teaming

Data Collection

Question Answering

Freelancer Overview

I am an internationally experienced General Practitioner with 7 years of clinical experience, currently practicing in Oman. Alongside my clinical background, I have formal, structured training in artificial intelligence, holding specializations in AI in Healthcare, AI for Medicine, and AI for Everyone from Coursera. I bridge the gap between complex engineering workflows and real-world clinical medicine. I specialize in medical RLHF (Reinforcement Learning from Human Feedback), prompt engineering, red-teaming clinical safety protocols, and auditing LLM diagnostic logic against global, evidence-based standards (NICE, SIGN, etc.). I am available for remote, asynchronous medical AI training contracts.

Labeling Experience

Medical LLM Safety Alignment & "Red Teaming"

TextRLHF

Focused on stress-testing and "red-teaming" Large Language Models (LLMs) to prevent medical hallucinations and prioritize patient safety. Designed complex, ambiguous primary care prompts (e.g., subtle presentations of myocardial infarction or atypical diabetic ketoacidosis) to evaluate model safety thresholds. Audited AI-generated responses using pairwise comparison, grading the outputs based on clinical accuracy, severity triage, and adherence to international medical guidelines (NICE/SIGN). Corrected high-risk medical errors and rewritten model outputs to ensure safe, evidence-based patient-facing communication.

2025 - Present

Clinical Chain-of-Thought (CoT) Verification

TextClassification

Evaluated and annotated the step-by-step diagnostic reasoning (Chain-of-Thought) of AI models processing Electronic Health Record (EHR) data and complex patient histories. Reviewed model logic for differential diagnosis formulation, pharmacological interactions, and diagnostic workups. Flagged instances where the model experienced logical drift or relied on structurally unsound clinical premises. Rewrote training data to teach the model how to properly synthesize history, physical findings, and lab results into a coherent, safe clinical plan.

2025 - 2026

Education

Work History