Eka Care has launched Parrotlet-a v2, India’s most performant clinical-grade, real-time Automatic Speech Recognition (ASR) model purpose-built for healtcare documentation, setting a new benchmark for AI-driven medical transcription across Indian hospitals and outpatient departments. Engineered specifically for the linguistic, acoustic and operational realities of Indian healthcare, the model enables doctors to generate structured clinical notes in near real-time, addressing the growing documentation burden accompanying rapid healthcare digitisation.
As electronic medical records (EMRs) scale nationwide, clinicians are spending increasing amounts of time entering structured data—often at the cost of patient interaction. In high-volume OPDs and hospital environments, manual documentation contributes significantly to workflow inefficiencies and rising burnout. Parrotlet-a v2 directly addresses this challenge by facilitating the transformation of doctor–patient conversations into structured, clinically accurate records within seconds, making near real-time documentation viable at scale rather than an added administrative burden.
“India’s clinical reality is multilingual, acoustically noisy, and filled with hyper-local medical terminology that global AI systems are not trained to handle,” said Vikalp Sahni, Founder & CEO, Eka Care. “Parrotlet-a v2 is tuned for Indian healthcare. In benchmark evaluations against leading global and India-focused models—including Gemini 3 Pro, Gemini 3 Flash and Saaras V3—it demonstrated leading Semantic Word Error Rate performance while delivering sub-second inference speeds, enabling documentation to happen seamlessly during consultations.”
The model powers EkaScribe, Eka Care’s AI-powered digital medical scribe platform, currently used by over 3,000 doctors. As a compact 5B parameter model, Parrotlet-a v2 bridges the gap between lightweight efficiency and large-model intelligence—matching and even outperforming larger models in Indian clinical settings while remaining economically viable for large-scale deployment. It is designed to interpret code-mixed Hindi and Indian English conversations, overlapping dialogue, regional accents and India-specific drug names while minimising hallucinations—instances where AI generates details that were never spoken.
Based on internal benchmarking across proprietary, real-world doctor–patient dataset, Parrotlet-a v2 achieved nearly 93% medical keyword accuracy in Indian English and 85% in Hindi, reflecting high precision in capturing critical entities such as drugs, dosages, symptoms and diagnostic terminology. In controlled evaluations on noise-only datasets, the model recorded one of the lowest Phantom Speech Rate of 3%, ensuring that ambient sounds are not incorrectly transcribed as clinical information—an important safeguard in high-noise hospital environments.
While some large general-purpose models can take up to 30 seconds to process a 30-second clinical recording, Parrotlet-a v2 delivers comparable accuracy with sub-second inference—representing up to a 30x speed advantage in real-world clinical workflows and supporting the 5–10 second documentation window required in busy OPDs.
“Healthcare digitisation at scale requires AI that is not only accurate, but fast, reliable and economically sustainable,” said Deepak Tuli, Co-founder and COO, Eka Care. “Our focus was to build a specialized AI engine that truly understands Indian clinical practice so providers can deploy it confidently across networks without compromising workflow speed, affordability or patient safety.”
With the launch of Parrotlet-a v2, Eka Care aims to support nationwide healthcare digitisation by reducing administrative overhead and enabling structured data capture at the point of care. The company will continue advancing India-focused health AI research, expanding language coverage and strengthening integrations with hospital, insurance and public health systems to power a more connected and data-driven healthcare ecosystem.
