ASR, LLMs, VAD, MFA, Python, Java, React, Django

He is a Data Engineer with over 1 year of experience. Currently, he is a Data Engineer at XXXX, where he serves as the lead architect for the company’s audio intelligence ecosystem, focusing on the end-to-end lifecycle of speech-to-text transformation. He deployed and fine-tuned multiple Automatic Speech Recognition (ASR) models. He leveraged prompt engineering and LLMs to correct Cantonese homophones and regional linguistic variations, elevating transcription quality to enterprise-grade accuracy. He designed a multi-stage audio processing engine capable of handling tens of thousands of hours of data. His pipeline integrates VAD (Voice Activity Detection), speaker diarization, noise reduction, and MFA (Montreal Forced Alignment). He also optimized batch processing through multi-threading and memory management, resulting in a 40% increase in processing speed and ensuring a normal distribution of audio segment lengths for model stability. And he integrated LLMs and signal-to-noise ratio (SNR) metrics to automate quality control, drastically reducing the reliance on manual annotations by filtering out low-quality audio at the source.

2.       What skill(s) / experience would you self-describe as strongest or specialist in?  

Strongest skills would be Full-Stack & UI/UX Development as well as data engineering.

3.       Do you have the lists of key criteria’s according to the clients’ requirements?

He is a Data Engineer with over 1 year of experience. He has professional experience building applications using the React and Django stack. He is also proficient in Python, Java, and C#/. He has deep experience in LLM prompt engineering and model optimization. He has specifically optimized transcription models for Cantonese and Mandarin, achieving nearly 90% quality. He has designed multistage processing pipelines involving VAD, noise reduction, and speaker diarization, utilizing multi-threading to increase speed by 40%. He is proficient with AWS (EC2, NGINX), Docker, and various databases including PostgreSQL and DynamoDB. He is currently managing the deployment and evaluation of ASR models and LLMs, processing over 20,000+ hours of audio data.