IndoML Multilingual Intent Detection — Rank 10

Hosted by IIT Bombay · November 2023 – December 2023

Prototype

Utterance

Pre-scored examples using real Amazon MASSIVE intent labels — the competition system was a fine-tuned BERT classifier over 52+ languages.

Executed large-scale multilingual text classification experiments on the Amazon MASSIVE dataset spanning 52+ distinct natural languages.
Partnered with a 3-person team to fine-tune pre-trained BERT sequence classifiers using Hugging Face modules, adding custom Word2Vec vectorization sequences and structured prompt engineering templates over open-source foundational models.
Secured Rank 10 out of all competing national teams, hitting a validation classification accuracy score of 90% and an official final leaderboard index of 0.86.