{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/1b6efdffa4634d44b3ed2eb392cf0a22\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/1b6efdffa4634d44b3ed2eb392cf0a22-07bb0e7c01279414.gif","duration":836.993,"title":"Building a Production-Grade Extraction Pipeline for Caller Information 📞","description":"Hi everyone, in this video, I walk you through my submission for the phone board challenge, where I built an extraction pipeline to gather caller information from German audio recordings. I utilized 30 WAV files and implemented two versions of the pipeline, achieving an overall accuracy of 92%, with notable improvements using prompt optimization techniques. I also incorporated observability features and a knowledge base for name grounding, but I identified areas for improvement, particularly in the speech-to-text model's accuracy. I encourage you to run the provided commands to explore the repository and test the different pipeline versions for yourselves. Thank you!"}