Sarvam is at the forefront of building sovereign AI for India, developing a comprehensive full-stack AI platform. We focus on research, models, infrastructure, and applications to make AI impactful for the nation. Backed by prominent investors and partnering with leading Indian brands like Tata Capital and CRED, Sarvam is dedicated to advancing AI's capabilities for India's growth.

This role involves owning key charters for Automatic Speech Recognition (ASR), Text-to-Speech (TTS), and model evaluation. You will collaborate closely with our expert research and engineering teams to shape product roadmaps, manage training and evaluation processes, and deliver models that significantly enhance user experience and system reliability.

Key responsibilities include defining and driving product roadmaps for ASR, TTS, and evaluation systems. You will partner closely with engineering and research to oversee data collection, training cycles, and model release strategies.

Establishing and managing evaluation frameworks is crucial, focusing on quality metrics such as accuracy, latency, Mean Opinion Score (MOS), and robustness. Running structured experiments, including offline tests, A/B testing, and human evaluations, will be vital for quantifying model improvements.

You will also ensure the scalable deployment and ongoing monitoring of models in coordination with infrastructure and data teams. Translating user and business requirements into actionable model goals and clearly communicating priorities, timelines, and trade-offs to all stakeholders are essential aspects of this role.

Candidates should possess a minimum of 3 years of experience in building AI, applied AI, ML products, or model training pipelines. A solid understanding of the complete model lifecycle—from training and fine-tuning to deployment and monitoring—is required.

Strong analytical skills and a proven ability to make data-driven decisions are essential. The ability to collaborate effectively across research, engineering, and product teams is also a key requirement.

Familiarity with speech technologies like ASR and TTS, or with evaluation pipelines, would be highly advantageous. Experience with quality metrics such as Word Error Rate (WER), Character Error Rate (CER), MOS, or latency tracking, alongside hands-on experience with SQL, Python, or ML toolkits, are considered bonus points.