Developed an inhouse speaker diarization system from the start using state-of-the-art models for contact centre data. It involved analysing the data, handling noisy labels and experimenting with different configurations to achieve the best results.

Worked on music retrieval and recommendation for multimodal design documents which contains both text and image data. It involved identifying the best feature vectors for image, text and audio and map them to a common space to find the most similar audio with the highest score. This was the first work which involved music recommendation for design documents and hence involved data curation and sampling task along with model development.