Sean Scholte Full Stack Developer

Info Support

Info Support

Developed an Augmented Multimedia Platform that automatically enriches uploaded podcast content with AI-generated textual content. The result was a web application where users can upload a podcast, after which the audio is transcribed and transformed into various text formats. The project started with research into existing AI systems for converting podcast audio into transcripts and generating derived text formats, such as summaries, titles, articles, tweets, quotes, and other content types. Based on this research, a web application was developed using Django and MariaDB, supported by two microservices built with Flask and Redis: one for transcription and one for generating additional text formats. The transcription service implemented the Whisper AI model, while the GPT-4 API was used to transform transcripts into different types of written content. Users could chain prompts in a tree structure (implemented with a task tree mechanism in celery).

Kadaster

Developed a Feedback Loop system for the automatic training and continuous improvement of AI models that predict and extract relevant information from official deeds. The system was built with Spring and Kotlin and integrates with Kadaster's object storage environment for deed documents. The application connects to multiple data sources, including one Oracle database and two PostgreSQL databases, to support validation, feedback processing, and data enrichment. This system contributes to improving the reliability and quality of AI-driven document processing by creating a structured connection between model predictions, stored deed data, and verified records.