Prompt & Content Designer: Conversational Speech Data

ARTPARK (AI & Robotics Technology Park) is India's first public-private partnership initiative focused on AI and robotics for social good, incubated at the Indian Institute of Science (IISc), Bengaluru. Established under the National Mission on Interdisciplinary Cyber-Physical Systems (NM-ICPS) of the Department of Science and Technology, Government of India, ARTPARK operates at the intersection of frontier research and real-world deployment.

As an innovation hub anchored at one of Asia's premier research universities, ARTPARK bridges the gap between academic research and impactful technology. Its programmes span healthcare, agriculture, mobility, and language, always with the goal of building solutions relevant to India's scale and diversity.

Project Vaani is India's largest and only multimodal Indic language speech dataset initiative, developed at ARTPARK in collaboration with IISc and funded by Google DeepMind. Its mission is to build a high-quality, linguistically diverse, and ethically grounded corpus of spoken language data across Indian languages and dialects: data that is critical for training and evaluating next-generation automatic speech recognition (ASR) and spoken language understanding systems.

Across its first two phases, Vaani has collected data from over 165 districts across 31 states and union territories, covering 109 languages, and recording more than 31,000 hours of speech from over 156,000 speakers. The data is open-sourced and available through the Vaani website and Bhashini, the Government of India's national language translation platform.

Learn more

• Project Vaani website: [vaani.iisc.ac.in](https://vaani.iisc.ac.in)
• Dataset on Hugging Face: [ARTPARK-IISc/Vaani](https://huggingface.co/datasets/ARTPARK-IISc/Vaani)
• Media: [ARTPARK-IISc launches Project Vaani, Analytics India Magazine](https://analyticsindiamag.com)

Phase 3 of Project Vaani focuses on collecting conversational speech data from languages and dialects that have been underserved by existing AI datasets, specifically targeting influential dialects with large speaker populations, as well as lower-resource languages where the need for quality speech data is acute and the ecosystem impact would be high.

The data collection methodology uses image-based prompts: a speaker is shown an image representing a scene or scenario and records their spontaneous spoken response. The quality of this corpus (its topical diversity, cultural representativeness, and responsible design) depends on how well the underlying prompts and scenarios are conceived. This is where this role sits.

We are looking for a Prompt & Content Designer to lead the design of conversational scenarios, topic sets, and image generation prompts for Phase 3 of Project Vaani. This is a 3-month contractual or internship engagement.

You will work at the intersection of social sciences, language, and generative AI, translating a deep understanding of India's sociocultural diversity into structured, balanced, and imaginative prompt sets that will shape what India's AI systems learn to understand.

• Design a comprehensive topic taxonomy for conversational data collection (covering domains such as agriculture, health, family life, local commerce, education, festivals, environment, gender and community), ensuring balance and representativeness across the target language communities
• Develop specific scene and scenario briefs for each topic, thinking through how different communities, genders, age groups, and geographies would engage with them
• Write image generation prompts (text-to-image) that visually represent each scenario in a culturally accurate and inclusive way
• Use LLMs (such as GPT or Gemini) to generate, refine, and evaluate prompt sets at scale: you will direct the model, not just use it
• Collaborate with ML engineers and researchers to ensure the prompt taxonomy aligns with the vocabulary and linguistic diversity goals of the dataset
• Navigate sensitive topic areas (e.g., health, gender, hygiene, social norms) thoughtfully, determining what can be included, what context is needed, and how to do it responsibly across different regional and cultural settings
• Document the prompt design framework so it can be reviewed, iterated upon, and extended

• Work on a dataset that will influence the trajectory of Indic language AI over the next decade, shaping how AI systems understand languages spoken by hundreds of millions of people, and impacting the entire Indic language technology ecosystem
• Collaborate directly with India's leading language and AI researchers at IISc, and with researchers from Google DeepMind, working at the frontier of voice AI and large language models
• Design the data collection framework from the ground up: your work will determine what the models learn, and by extension, how well they serve millions of speakers of underrepresented languages
• For social science backgrounds: a rare and direct pathway into applied AI and prompt engineering, with real project credits and hands-on exposure to how data shapes frontier AI systems
• For AI/data backgrounds: deep exposure to responsible data design and sociolinguistic complexity that most practitioners never encounter

Apply using the form: forms.gle/FWRJDMjnxPYubhWF8. In the form, you will be asked to briefly describe one example of a time you worked with an LLM to get a structured output, and one example where you navigated a question of representation, bias, or cultural sensitivity in any context.

Applications will be reviewed on a rolling basis.