Multilingual question answering for media

I implemented an automated multilingual question-answering system for DosBro’s media platforms, letting users query dynamic content in three Indian languages instead of relying on static search or manual curation.

Built with PyTorch and attention-based architectures, the system handled the messy reality of user phrasing across languages and increased user engagement by 30%. The hard part was not the model architecture alone but making QA feel responsive on live, updating content.

Focus

• Multilingual QA across three Indian languages
• Attention-based PyTorch models on dynamic content
• User-facing NLP with measurable engagement lift