Berater, AI Engineer und Softwareentwickler

This project explored enhancing question answering within lecture videos. For that purpose, we developed a multimodal Retrieval-Augmented Generation (RAG) framework. Recognizing the growing abundance and popularity of educational videos online, we tackled the challenge learners face in efficiently locating relevant information within these videos.

Our framework integrates advanced RAG techniques, leveraging both audio and visual data through automatic speech recognition (ASR) and optical character recognition (OCR), to extract context and enhance the accuracy of generated answers. The framework significantly outperforms traditional standalone language models, demonstrating the effectiveness of multimodal retrieval combined with context re-ranking.

The work has been published in Springer's Lecture Notes in Computer Science.

Psifoperma is a derivation of the Greek word psifos (en: vote) and perma for permanence.

The purpose of the project was to build a voting/survey system for clubs and small companies, using the ERC-20 token standard on the Ethereum blockchain. Developed with Meteor.js, it offers secure and tamper-proof voting records, immutability, transparency, and pseudonymity. Users can create votes and invite voters, voters can easily cast votes, and both users and voters can view the results.

The app provides direct links to Etherscan, which makes each transaction via the smart contract (creation of votes, casting votes) transparent and public. The deployed smart contract can also be used directly, without our application. Furthermore, the contract technically already allows voters to vote directly (either the vote creater or the voter can cast a vote for the voter).