GSoC 2026 Fine-tuning AI Transcription for Greek Municipal Councils

This project fine-tunes a Whisper ASR model using LoRA to enhance the first stage of OpenCouncil's transcription pipeline. Greek municipal speech breaks standard Whisper in predictable ways: council-specific idioms, proper names it has never encountered, and morphological errors that accumulate fast in a heavily inflected language. The goal is a 15–20% drop in domain-specific Word Error Rate (DS-WER). The expected result is that human reviewers spend less time correcting transcripts and more time on work that actually requires judgment.