A team of software engineers created a prototype AI system called Klartext to analyze political speeches in real-time at Sweden's Almedalsveckan. They faced challenges like live speech transcription and prompt engineering for Large Language Models. Key Performance Indicators (KPIs) such as promise tracking and emotional vs. rational scoring were measured. The system's back-end was built to scale using Firebase's cloud functions and maintained real-time updates through Firestore. The prototype successfully transitioned to a production-ready state, providing insightful real-time analysis during the event.
Table of contents
Building a prototypeWhat do we want to measureHow do we measureListening to the speechProcessing the speechDemonstrating the prototypeTechlead / senior software engineerAllround back-end developer.NET back-end developerAndroid-developerJunior B2B-contentmarketeerOpen sollicitatieStageReal-Time Transcription and AI Analysis in ProductionTowards an event-driven cloud architectureIntroducing Speech Configurations and Event StatesFurther optimizations: finding the right tools for the right tasks, at the right timeLessons Learned and Looking AheadSort: