A structured catalog indexing ongoing, completed, and upcoming NLP research projects conducted jointly between the Department of Artificial Intelligence (DoAI) and the Information and Language Processing Research Lab (ILPRL).
A unified, audit-first data operating system built on Datasette and SQLite. Ingests raw web crawls and archive dumps, normalizes legacy fonts (Preeti/Kantipur), filters Hindi leakage, clusters MinHash duplicates, and runs native-speaker annotation queues with complete provenance tracing.
A rigorous empirical evaluation benchmark and baseline suite testing generative reasoning, instruction-following, and natively multimodal retrieval across 22 years of Nepali computational linguistics assets, pairing Gemma 4 E4B with EmbeddingGemma 2.
Investigating end-to-end Conformer/Whisper fine-tuning on open speech archives (OpenSLR-54), time-aligned transcript verification, and dialectal recording campaigns for under-represented languages of Nepal.
Developing high-accuracy vision-language layout parsers and algorithmic Preeti-to-Unicode converters for historical Nepal Gazettes, Supreme Court rulings, and national curriculum archives.
