Jugalbandi Studio: Bridging the gap between entrepreneurial vision and AI know-how
OpenNyAI, a collaborative mission founded by Agami, EkStep Foundation, Thoughtworks, Rohini Nilekani Philanthropies and National Law School of India University is deeply involved in developing open AI public goods for high-impact applications in the law and justice space in India.
Last year, on the heels of the launch of ChatGPT, OpenNyAI created Jugalbandi, a free and open software stack that combines the power of ChatGPT and Indian language translation models under the Government of India’s Bhashini mission to power conversational AI solutions in any domain.
As a result of what we heard and what we sensed this year, OpenNyAI collaborated with Microsoft Research India, to develop and release Jugalbandi Studio to enable innovators to bring the power of Jugalbandi to their solutions easily. The Studio enables -
An open source environment which allows organisations to articulate their chatbot's flow using natural language (Simply say, “Provide information to user on the government schemes they may be eligible for after cross questioning them on the three relevant criteria to determine their eligibility.”)
Beyond information extraction, it enables actions like digital payments, e-signatures, Aadhaar authentication, and form filling (“Once the user has confirmed which scheme they want to apply for, ask them for the relevant details and send them a filled form in PDF format. After this, you will need to do identity verification using Aadhar authentication.”)
Organisations can thus iterate rapidly on user flows and logics without the need for technical expertise or large capital investments. (“No, that is not what I meant. You have to only do Aadhaar verification once you have gotten the details for the form.”)
Most, if not all, problem statements within law and justice require the ingestion of sensitive information and thereby, careful handling of the information to generate responsible outputs. In the backend, this increases the requirements from the code to ensure that common and potential (also not so common) pitfalls are addressed.
Considering that these populations are especially vulnerable (people with low literacy, legally vulnerable communities, etc.), personally identifiable information (“PII” like phone number, address, Aadhaar number, etc.) is handled very sensitively in this stack - it cannot be made visible to any unauthorised persons and it cannot be exported to any external computer. The stack has been optimised to include best practices of generative AI - creating redundancies for bot-generated information, ensuring human-in-the-loop systems and explicitly filtering and masking PII.
The Studio engages with the programmer of the application (who does not need any pre-requisite technical skills) to point out such pitfalls and errors in the application and helps them fix these errors. For example, if the application's flow involves making a UPI payment, and the programmer forgets to handle the edge cases where the payment does not go through, the Studio reminds them about this and lets them engage in a dialogue to fix the issue. These capabilities enable the programmer to develop more robust applications for these vulnerable populations.
The Studio allows the programmer to articulate the flow of the chatbot, and interactively flesh out the details of every step, which could involve invoking various APIs. When all error cases are handled and appropriately addressed, it shows "No errors" in the middle pane (see image above).
Once you have articulated the flow of your chatbot, you can export it to your cloud-hosted version of Jugalbandi and within a few hours, have it ready for testing by your target audience on Telegram / WhatsApp. This allows an organisation to have control and visibility over the bot hosted by them and how they will be using it along with the data flowing in from the users. The technology powering Jugalbandi Studio is built by Microsoft Research and is called Programming with Representation (PwR). You can read more about PwR here.
To set context, in December 2022, three days after the launch of ChatGPT, Jugalbandi was born. And with it was born the possibility of finally bridging the information gap in Indian languages at population scale. It has been a year since organisations across for-profit, not-for-profit, government, and academia have come together to imagine a world where each Indian is aware of their rights and has a pathway to successfully acquiring those rights. Decades of ongoing issues like complex government scheme application processes, low awareness of legal rights, inaccessible systems of grievance redressal, and systems designed to render citizens without any agency - required us to think bigger and better. And AI helped.
In the past year, we have continued to build the underlying technology of Jugalbandi - a software stack that makes it possible for anyone to gain access to reliable information in their own language and in voice (to those uninitiated, India has 22 official languages) on a medium of their choice (WhatsApp being the popular choice) in real-time. The Bhashini mission continues to expand the number of Indian language models. In this duration, we have consulted and been approached by close to 100 organisations (across for-profit, not-for-profit and government) to use the power of Jugalbandi to solve the problems inherent in their organisations, present amongst their audience etc.
Arriving at Jugalbandi Studio has been a journey of working with these organisations that were trying to solve an array of problems. These ranged from solving for internal workflows, providing financial literacy, agricultural best practices, legal information and legal aid, access to and information on government welfare programs, conducting e-commerce search, grievance redressal, access to housing for migrant workers, providing e-education to students, mapping hotspots of environmental initiatives, upskilling of illiterate and uneducated citizens, creating field notes and documentation of success stories of NGOs, storytelling assistance to capture grassroots stories, etc.
There was also a wide category of users - from direct beneficiaries, to on-ground volunteers of CSOs, students, legal teams and law firms, judges, entrepreneurs and bureaucrats.
We spoke with organisations to understand where Jugalbandi can add value to their workflow and culled out some decision-making criteria to determine the same across organisations -
Do they even need an AI-based solution, or does it require a human-centric solution?
Technology capability within the team/capacity to use, make sense of and deploy technology within the organisation
Clarity of use-case and problem definition
Willingness to adopt AI and do thorough testing for responsible application
During these conversations, and after many reflections, we identified a few major roadblocks in deploying technology/AI/Jugalbandi for private and public organisations alike - one, the challenge of understanding how the technology works and what its capabilities are. For instance, if a bot has the ability to provide information on government schemes, the same underlying logic can work to enable a storytelling bot that captures field stories of NGOs. Two, there is a lack of necessary resources to actually deploy the tech for your user base at scale. And three, managing the expectations of the users as soon as information is provided since the expectation is also for the bot to do the next steps/perform certain tasks once the information is rendered.
We launched Jugalbandi Studio at the Agami Summit in December 2023 and are soon going to start a pilot program with 10 organisations to identify even more action-based features (like e-KYC, digital payments, etc.) that can be incorporated for plug-and-play even as we work toward open sourcing the platform in the coming few days. We hope that 2024 will be the year that hundreds of innovators are able to bring their voice-powered justice solutions to the field thanks to Jugalbandi Studio.
 | A guest post by
|