About Deepon Halder

I am a machine-learning researcher from India working across automatic speech recognition, multilingual natural-language processing, language models, and low-resource language technology. I currently work on improving ASR models at Alma Inc. My earlier research and engineering experience includes Approximate Labs, Cisco, and AI4Bharat at IIT Madras. I am also an undergraduate at IIEST Shibpur.

My research focuses on making language and speech systems more useful across languages, cultures, and resource settings. Recent work includes large-scale preference evaluation for Indian-language text-to-speech, adaptive sampling for autoregressive language models, synthetic Indic-language story corpora, cultural calendar reasoning, transliteration in NLP, generative reasoning benchmarks, and cyclical distillation for machine translation. Publications and primary-source links are listed on the portfolio homepage.

This site is my canonical public portfolio. It collects my research, open-source projects, academic service, talks, and technical writing. The blog archive contains practical explanations of machine learning, NLP, speech systems, language models, and computer systems. For collaboration, corrections, speaking invitations, or other professional questions, use the contact page.