A two-stage intelligent dialogue system for Assam tourism, featuring a MuRIL-based intent classifier across 44 intents with ~97% accuracy and a semantic retrieval module over 221,799 Q&A pairs covering 51 destinations - designed for low-resource code-mixed NLP.
A research-grade NLP dialogue system for Assam tourism that handles code-mixed Assamese-English queries — a virtually unexplored frontier in low-resource language AI. This system bridges the gap between how locals actually speak and how machines understand language.
Code-mixed Assamese-English (mixing both languages within a single sentence) is spoken by millions but almost entirely ignored by NLP research. No existing system could handle Assam tourism queries in this natural code-mixed form across the diversity of 51 tourist destinations.
Built a two-stage pipeline: first, fine-tuned Google's MuRIL (a multilingual BERT variant pre-trained on Indian languages) for intent classification across 44 distinct tourism intent categories. Second, developed a semantic retrieval module that searches over a curated corpus of 221,799 Q&A pairs covering 51 Assam tourist destinations to return accurate, relevant answers.
I'm available for freelance projects. Tell me what you have in mind.