Muhammad Baha'udin Mursid

I’m an AI training and evaluation specialist with 3+ years of remote experience across search relevance (Microsoft Bing/UHRS), LLM fine-tuning and RLHF (Google Bard/DataCompute), and speech-to-text (ASR) evaluation—consistently delivering 89%+ accuracy.

As an AI Trainer at Tech Mahindra (remote), I fine-tuned large language models for NLU/NLG tasks on the Google Bard project, producing human-labeled training data to improve accuracy and contextual relevance. I achieved approximately 89% accuracy on RLHF evaluations using the DataCompute platform, while analyzing model outputs against quality guidelines and flagging inconsistencies and edge cases for retraining.

In my search engine work with TELUS International (remote), I evaluated and rated search engine results for relevance, accuracy, and alignment with user intent on UHRS (Microsoft Bing evaluation platform). I completed 5,000+ submissions per task type across 100+ distinct task types and achieved an average accuracy rate of 91.5%, delivering structured feedback used directly by engineering teams to refine search ranking algorithms.

Most recently, as a Speech-to-Text Evaluator on the Echo Project (RWS, remote), I evaluated speech-to-text model transcriptions for accuracy, fluency, and alignment with source audio across speaking styles and accents. I identified and documented transcription errors and model failure patterns to support ASR improvement, applying structured quality guidelines to ensure consistent, reliable human-labeled evaluation data—backed by a strong English Literature foundation in analytical writing and content judgment.