23/07/2025
Thrilled to share our AI/ML internship project at eSewa→ BhasaBODH 🇳🇵
Our goal was to build a Nepali language understanding model that worked well with both Devanagari and Romanized scripts. Using BLOOMZ-3B and QLoRA, we fine-tuned a multilingual model to handle tasks like both transliterations (Romanized to Devangarik & Vice-Versa) and chatbot interactions. We worked with 90K+ lines of data, ensured data privacy, and deployed the model using FastAPI with a web interface. It was a great hands-on experience in model training, evaluation, and integration with real-world systems.
Project Report: https://drive.google.com/file/d/1EbmsxLG8mnIju_PzJwk6xqXoVgpSu8qG/view?usp=drive_link
YouTube: https://youtu.be/yEzuXaCNi-4?si=3B_7_uOQeXIXeYyd