About the job
About Anuvaya
At Anuvaya, we envision a future where conversational AI agents will revolutionize professional services across India, starting with astrology. Our dedicated team of engineers, designers, and product specialists is committed to blending conversational AI with domain-specific knowledge. Crafting an AI agent that sounds human is challenging, and achieving domain expertise adds another layer of complexity. We're tackling both challenges simultaneously.
Supported by industry leaders like Accel, Arkam Ventures, and Weekend Fund, we are on the forefront of innovation.
The Role
Our Applied AI Research team is focused on enhancing the performance of conversational agents in Indic languages. Frontier models excel in English but struggle with languages like Hindi, Hinglish, Tamil, and Telugu. Our primary goal is to ensure our AI agent communicates fluently and accurately across these languages, achieving expert-level understanding.
In this role, you will be pivotal in bridging the gap between research and practical application. You will conduct finetuning experiments to enhance response quality in languages, redesign context management for multi-turn conversations, and develop robust systems that empower our models to act as domain experts.
What You'll Do
- Finetune models to enhance fluency, accuracy, and domain knowledge in Hindi, Hinglish, and other Indian languages.
- Develop and oversee finetuning pipelines, including data curation, training runs, evaluation, and deployment of refined models.
- Collaborate with the team on prompt engineering and context management to optimize how models process information in multi-turn conversations.
- Design retrieval strategies to ensure timely access to relevant domain data.
- Conduct experiments to analyze how variations in finetuning, context structures, and prompt formats impact output quality across languages.
- Work closely with the Evaluation team to assess critical quality metrics, particularly in subjective and language-dependent contexts.
- Stay updated on the latest advancements in frontier model capabilities and find ways to leverage new features as they become available.
