What Medharvix builds.

Machine translation systems for low-resource and underserved Indian languages. We fine-tune multilingual foundation models on curated parallel corpora, evaluate against native-speaker benchmarks, and deploy production-ready translation pipelines.

Our flagship work in Khasi machine translation (Bha-Kha V4, BLEU 48.0) demonstrates this capability through Bhasaflow, a Medharvix Systems platform. We are expanding to additional Indian languages using the same evaluation-led approach.

Text-to-speech and automatic speech recognition systems designed for the acoustic, phonological, and prosodic patterns of Indian languages. Our speech work focuses on languages where existing commercial TTS and ASR systems provide poor or no coverage.

Khasi TTS is in active research. Khasi ASR is under development. Both are designed to integrate into the Bhasaflow platform stack.

Optical character recognition and document understanding systems for printed and handwritten material in Indian scripts. This capability enables digitisation, searchability, and preservation of institutional and historical records.

Khasi OCR is under active research, with a focus on reading both typed and handwritten Khasi text from scanned documents.

Systematic model fine-tuning on open multilingual foundations (NLLB, Whisper, and others), with structured evaluation pipelines that measure real-world performance — not just training loss. We build evaluation infrastructure as a first-class capability.

Structured data collection, cleaning, annotation, and corpus preparation workflows that support training and evaluation of language, speech, and document models. We work with native speakers and community annotators to ensure data quality and linguistic accuracy.

Design and development of integrated AI platforms that bring multiple capabilities — translation, speech, document understanding — into unified, deployable, and maintainable systems. Bhasaflow is the primary example of this platform capability.

Workflow automation and intelligent decision-support systems for institutional and enterprise use cases. This capability focuses on applying AI to operational problems where structured automation can deliver measurable efficiency gains.