Simplified language-data contribution for Bhashini
I made voice, text, translation, and image contribution easier to understand and complete.
Role
UX/UI Designer | AI contribution and accessibility
Duration & company
Add confirmed project dates and duration
My Squad
Design, Product, Engineering, and Mission partners
Context
AI could not support India’s languages without data from the people who speak them
India has more than 120 languages, but most digital services still work best in English. Bhashini was building public language datasets so AI-powered products and government services could work for more people in their native language.
The mission depended on voluntary contributions. People could record speech, validate audio, write or translate text, and contribute image data. The design challenge was turning a large national technology program into a task an individual could understand and complete in a few minutes.
Problem
The mission mattered, but the contribution experience asked too much from first-time users
People had to understand Bhashini, learn why language data mattered, choose between four contribution modes, understand a new task, and trust that their work would be useful.
The platform explained the program, but it did not make an individual contribution feel immediate or meaningful. Voice recording, validation, translation, and image tasks also behaved like separate experiences, increasing the amount users had to learn.
The real problem was not simply usability. It was motivation. People needed to know why their contribution mattered, what to do next, and whether their small action made a difference.
Research
The research reframed participation around purpose, confidence, and visible progress
I reviewed how the existing experience introduced the mission and how people moved across Suno, Bolo, Likho, and Dekho, the four modes for audio validation, speech, writing or translation, and image contribution.
The friction clustered into three patterns. People could not quickly connect the national mission to one small task. Instructions appeared before users had enough context to choose confidently. After contributing, the impact remained invisible.
I used those patterns to shift the product from explaining a platform to helping someone make one useful contribution.
Decisions and tradeoffs
The mission needed to feel personal before it felt technical
Decision:
I changed the entry experience to lead with the human value of language access, then introduced contribution choices as small actions people could take immediately.
Tradeoffs:
Reducing the upfront explanation meant some institutional and technical context moved deeper into the experience.
Constraints:
The platform still needed to communicate credibility and explain how public contributions would support national language AI.
Result:
People could understand the purpose before being asked to learn the mechanics of contribution.
Four contribution modes needed one shared mental model
Decision:
I used a consistent flow across Suno, Bolo, Likho, and Dekho: understand the task, see an example, contribute, confirm the result, and continue.
Tradeoffs:
A shared structure reduced learning effort, but each data type still needed specific instructions and validation rules.
Constraints:
Audio, text, translation, and image contributions could not use identical controls or quality checks.
Result:
Users could move between contribution modes without relearning the entire product.
Instructions worked better at the moment of action
Decision:
I reduced the amount of explanation shown before a task and moved short instructions, examples, and validation feedback closer to the action they supported.
Tradeoffs:
Showing less upfront created a faster entry point, but required stronger contextual help inside each task.
Constraints:
The experience had to remain understandable for people with different levels of digital literacy and language fluency.
Result:
People could begin contributing without first understanding the entire AI data ecosystem.
Progress had to make an invisible contribution feel real
Decision:
I introduced clear completion feedback, personal progress, and community contribution signals to connect one small task to the larger mission.
Tradeoffs:
Gamification could improve motivation, but too much competition would undermine the civic and inclusive purpose of the platform.
Constraints:
A recorded phrase or validated translation does not produce an immediate visible outcome for the contributor.
Result:
Completion and community progress gave contributors a clearer signal that their work counted.
Contribution quality mattered as much as contribution volume
Decision:
I used examples, confirmation states, and task-specific feedback to help people understand what a valid contribution looked like before submitting it.
Tradeoffs:
More validation improved dataset quality but added friction and could discourage first-time contributors.
Constraints:
The interface needed to support useful AI training data without expecting contributors to understand technical quality standards.
Result:
Contributors received clearer guidance while the platform protected the usefulness of submitted data.
Outcome
A clearer reason to participate
The experience connected language contribution to better digital access instead of leading with the technology.
One interaction model across four activities
Consistent task stages reduced the amount contributors had to relearn.
More confidence during contribution
Examples, contextual instructions, and validation clarified what a useful submission looked like.
Visible progress connected individual work to the mission
Completion feedback and community signals made an otherwise invisible contribution feel meaningful.
Behind the scenes
Process: Existing-experience review, behavior mapping, persona and journey synthesis, contribution-flow design, prototyping, usability critique, and iterative visual design.
Scope: Responsive contribution experiences across voice, audio validation, writing, translation, and image tasks.
What I would measure next: Start-to-completion rate by contribution type, abandonment points, repeat contribution, quality acceptance rate, and participation by language.
What I learned: People do not participate in a mission because the technology is impressive. They participate when the purpose is clear, the task feels achievable, and they can see how their small action contributes to something larger.
