Language Mismatch Detection with LLMs

Context: Built as a fallback mechanism for a multilingual AI platform serving 80+ countries.

Challenge: Detect mismatches between text content and associated audio files to ensure language consistency and prevent user confusion.

Solution: Developed a Streamlit-based application using:

• LangDetect – for automatic language detection of input text
• Whisper – for transcribing audio files and comparing languages
• Built a full‑stack POC with OpenAI model, AWS S3, and MongoDB for phoneme generation

Outcome: Successfully integrated as a fallback layer, reducing language‑related errors and improving user trust across the main platform.