Active Research
Khasi TTS
Speech synthesis in active iteration — including studio-grade recording collaboration with Radio Punjab for smooth takes, clean frequency response, and natural voice quality in training data.
Flagship Platform by Medharvix Systems Private Limited
Language intelligence for the next billion.
Bhasaflow builds AI access for underserved Indian languages, starting with Khasi. We turn language research into real systems for translation, speech, OCR, and open resources.
Khasi-first. Research-driven. Built for real public use.
About Bhasaflow
Built for low-resource and tribal language AI at platform scale.
Bhasaflow is a research-led language AI platform focused on practical impact for communities that are typically left behind by mainstream models. It bridges languages, people, and AI systems through deployable translation models, speech technology research, OCR pipelines, and open language resources.
The long-term vision is clear: AI useful for every Indian, without language barriers.
Flagship Model
Bha-Kha V4
Flagship Khasi machine translation model, fine-tuned on NLLB.
Production Trajectory
Khasi-English MT
Research-Led Deployment
48
BLEU Score on Khasi MT
NLLB
Fine-tuning Foundation
V4
Current Flagship Generation
Research and Traction
Serious language R&D with measurable outcomes.
- Khasi machine translation and Khasi speech technologies under active development.
- Khasi TTS model work completed with continuing research iteration.
- Bha-Kha V4 MT stack fine-tuned on NLLB with BLEU 48 performance.
- 20 Khasi working collaborators across the language ecosystem.
- Under second-phase evaluation in the BHASHINI Khasi challenge as Team Techno Tuners.
Datasets and Resources
Open language assets published through Hugging Face.
bhasaflow-khasi-english-parallel-sample-v1
Curated sample set for rapid experimentation and evaluation of Khasi-English MT.
Open Dataset ↗bhasaflow-khasi-english-parallel-corpus-v1
Parallel corpus for model training, alignment tasks, and translation benchmarking.
Open Corpus ↗bhasaflow-khasi-monolingual-corpus-v1
Monolingual Khasi corpus for linguistic modeling and downstream language tasks.
Open Corpus ↗Live Product Experience
Use the Khasi-English translator now.
Experience Bhasaflow translation in production through the official Hugging Face Space.
Open Bhasaflow TranslatorSpeech and OCR Roadmap
From translation to full language infrastructure.
Launching 7 June
Khasi ASR
Automatic speech recognition stack for Khasi is nearing release.
Launching 25 June
Khasi OCR
Optical character recognition pipeline under focused development.
Recognition
Phase-2 evaluation in BHASHINI Khasi challenge.
Bhasaflow is under second-phase evaluation as Team Techno Tuners, validating real-world relevance of our Khasi-focused language systems.
Ecosystem
Research collaborators, institutions, and contributors around Bhasaflow.
Team Techno Tuners, Medharvix Systems, academic partners, and regional collaborators together shape this language AI ecosystem.
Student collaborators contributing as individuals — not a formal university partnership or endorsement.
- Team Techno Tuners in BHASHINI Khasi challenge (phase-2 evaluation)
- 20 Khasi collaborators and language contributors
- Radio Punjab collaboration for TTS-oriented recording — smooth sessions, high-quality capture bandwidth, and voice tracks tuned for clarity in synthesis pipelines
- Published datasets and tools through Hugging Face
Institutional and Media Collaborators
Powered by a growing ecosystem.
With Radio Punjab, we collaborate on text-to-speech data and recording workflows: controlled studio conditions, smooth session flow, and capture that preserves high-frequency detail and warm voice character — so models learn from audio that sounds good in the real world, not just on paper.
Build Language Equity With Us
Join the Bhasaflow journey for inclusive Indian language AI.
For innovators and contributors, use the dedicated contact page — forms save to Firebase once configured.