Established in 2021, OpenNyAI is a community of legal professionals, technologists, entrepreneurs, policy practitioners, students among others, with a common vision to advance AI for Justice in India. This community is anchored by Agami, EkStep, Microsoft, Thoughtworks & National Law School India University, Bangalore. By enabling creation of the necessary tools, infrastructure, and testbeds, OpenNyAI is making it easier and more affordable for organisations to demonstrate proof of concept across various knowledge bases, extending it beyond just law and justice.Driven by the OpenNyAI initiative, Jugalbandi's mission is to enable social entrepreneurs to harness AI for social impact.
Jugalbandi, at its core, is an information and action tool.
In today's information age, access to justice relies on the availability of quick, accurate, and verified information—be it regarding rights, entitlements, policies, or laws or any other matter that significantly impacts our lives. Yet many times, the intricate language and interconnected nature of such information scattered across various documents, departments, and institutions makes it intimidating for anyone to access it without formal assistance or training. Even those with formal training may struggle, losing precious time navigating the complex landscape to find the right information. Overall, this results in significant financial, opportunity, and time costs for everyone involved.
Institutions around the world, both private and government, have tried to tackle these issues across various domains, aiming to provide accessible and actionable information to all. However, these efforts are often limited in terms of both reach and resources, leaving many without the benefits they can potentially offer.
More importantly, information is only as powerful as the actions it can enable/inspire; without an added capability to act on the provided information, its potential remains unrealized. This is where Jugalbandi comes into play. Jugalbandi offers the potential of a transformative AI technology stack, leveraging conversational AI to make information on anything and everything - be it policies, laws, rights, or simply information required to navigate any task, both accessible and understandable. This stack allows the user to ask any query directly through an interface such as WhatsApp, in their native language. For instance, a user can ask a question in Malayalam using their voice and receive a response in Malayalam voice, making the interaction as natural as speaking to a human assistant.
Jugalbandi doesn’t stop at providing information. If a citizen wishes to take action based on the answer to their query, such as booking an appointment with a service provider or even initiating legal proceedings, such as filling out a legal notice or complaint, they can successfully do so through the same WhatsApp interface. Any entrepreneur, working in any field, can leverage this stack to build their conversational and action services at scale. Developers can pick and choose the components of Jugalbandi most suitable for them to build or improve their web apps, mobile apps, IVRS, etc. We will detail out these components in sections to follow.
Empowerment: This playbook serves as a comprehensive guide to Jugalbandi, enabling both functional and technical professionals from various backgrounds to grasp its features, components, and potential applications.
Deepen understanding: We also aim to engage people towards a deeper understanding of the capabilities, challenges, ethical considerations, and opportunities surrounding not just Jugalbandi but also other similar conversational AI stacks, particularly in the context of social impact initiatives.
Innovation: By democratising the know-how of Jugalbandi's capabilities, we aim to inspire professionals to explore innovative use cases and integrations, fostering continuous innovation on Jugalbandi.
Collaboration: Encouraging collaboration across different domains, the playbook facilitates knowledge sharing and collective problem-solving to maximise Jugalbandi's impact.
Ecosystem Expansion: Equipped with essential knowledge and tools, professionals can contribute to expanding the current Jugalbandi ecosystem, attracting more stakeholders to further enhance the platform.
Accessibility: it's not just the Jugalbandi stack that is accessible, the design principle cuts across all our efforts - documentation, benchmarking & dataset creation.
The current approach to AI-based solutions has two issues: the proprietary nature of the tools being developed and the lack of collaboration among creators. This results in AI solutions that require considerable private investment and are often kept behind closed doors. As a result, entrepreneurs, particularly in the development sector often start from scratch, leading to inefficiencies and higher costs. To overcome this, OpenNyaAI mooted the design and development of AI public goods like baseline models, datasets, and benchmarks, which, despite being expensive and time-consuming to develop, offer high repeat value. These public goods can be used to quickly build POCs that can aid any social entrepreneur to garner funding and support for various applications.
While the Jugalbandi stack is not restricted to any particular sector or domain, it was initially devised to make law, justice and government schemes more accessible to Indian citizens. With the low proliferation of legal services in the hinterlands, with some reports suggesting that only about 0.5% of the population having access to legal aid, this was the ideal domain as a unit of change. But the original proposition of accessing critical information in a simplified manner in your own language was powerful enough to not be bound by only a specific sector. Additionally, as the capability of the LLMs kept improving exponentially, it became increasingly clear that the applications of Jugalbandi that utilised these LLMs could be sector agnostic.
Although LLMs like GPT have been advancing rapidly from GPT-3 to newer versions like GPT-4, and GPT-4o, their quick development rate poses a challenge. While the digitally equipped can fully utilize these technologies, people lacking digital and language skills risk being left behind, widening the gap between those who are digitally literate and those who aren't. Additionally, these LLMs many times run the risk of hallucination which is an unaffordable flaw when providing with crucial information that directly impacts the actions that a user may take, Jugalbandi’s value add to address this problem is three-fold:
Make the conversational interfaces accessible across many more Indian languages
Restrict the possibility of the LLMs hallucinating. Overcoming a key roadblock to ensure that AI based solutions are more reliable for someone without the time or capability to immediately fact check the information provided to them
Through the use of finite-state machines and support across multiple channels, make various services available to the user beyond conversing with the LLMs, on any medium of their choosing
Jugalbandi is a versatile, scalable, and secure framework that democratises access to actionable information across languages, integrates with multiple LLMs, and ensures affordability and accuracy, promoting widespread adoption and innovation.
Accessibility Across Languages: In diverse regions with numerous spoken languages, breaking down language barriers is important to democratise access to critical information effortlessly.
Actionable Information: Information in itself is only valuable if it empowers users to take meaningful actions. By integrating capabilities to perform actions like booking appointments or initiating legal proceedings directly through the interface, Jugalbandi ensures that users can act on the information provided, thus maximizing its utility and impact.
Platform-Agnostic Framework: Users interact with technology through various platforms. Being platform-agnostic allows Jugalbandi to be deployed across multiple channels such as WhatsApp, Telegram, mobile apps, and web portals, ensuring that it meets users on their preferred platforms, improving user engagement and adoption.
Scalability: High user demand can strain systems, but Jugalbandi Manager is designed to handle a wide range of users, from just 10 to over a million. This scalability ensures that the framework can be effectively utilized by both small and large user bases, adapting to various levels of demand. The only limit is the infrastructure provided by the adopter of the stack (such as cloud services and the ability to incur any costs imposed in the usage of LLMs), making it a versatile solution for diverse organizational needs.
Integration with Multiple LLMs and Services: Different scenarios require different AI capabilities. By supporting integration with various language models like GPT, Llama, and Phi, as well as services like Bhashini, Azure, and Google, Jugalbandi offers flexibility and ensures that developers can choose the most suitable tools for their specific needs.
Accuracy and Reliability via Retrieval-Augmented Generation (RAG): Ensuring the accuracy and reliability of information is crucial, RAG in Jugalbandi helps in generating responses based on verified data from the knowledge base, minimizing the risk of misinformation and enhancing user trust.
Security and Privacy: Protecting user data is paramount. Features like sensitive information redaction, encryption, and controlled AI model access ensure that user data is secure and private, maintaining confidentiality and fostering trust among users.
Ease of Deployment and Customization: To encourage widespread adoption, Jugalbandi is designed to be easy to deploy and customize. This ensures that organizations and entrepreneurs, regardless of their technical expertise, can quickly implement and tailor the framework to suit their specific requirements, thereby accelerating innovation and impact.
Support for Finite-State Machines (FSM): FSMs provide structured and predictable conversational flows, essential for handling complex interactions and ensuring smooth user experiences. This design principle allows Jugalbandi to manage conversations effectively, providing users with coherent and logical interactions.
Open Source with Community Support: Open-source availability ensures transparency, fosters collaboration, and accelerates innovation. By providing the Jugalbandi code on GitHub, the framework invites contributions from a diverse community of developers, legal professionals, and social entrepreneurs, enriching the ecosystem and driving continuous improvement.
Affordability and Cost-Effectiveness: High costs can be a barrier to adoption, especially for social impact projects. Jugalbandi aims to be affordable by utilizing open-source components and minimizing additional charges. This ensures that even organizations with limited resources can leverage the framework to drive positive change.
Resilience Against AI Hallucinations: Ensuring the reliability of AI-generated responses is crucial, especially in high-stakes scenarios. By incorporating mechanisms to reduce the risk of hallucinations in LLMs, Jugalbandi ensures that users receive accurate and trustworthy information, enhancing the framework's credibility and effectiveness.
Jugalbandi leverages pre-trained models via RAG, extends beyond government schemes and chatbots, is platform-agnostic, requires customization, and offers free code but may contain some cost-bearing components.
As Jugalbandi continues to evolve and expand its reach, we think it’s helpful to clarify some common misconceptions to understand its true potential and capabilities better. Here are the key clarifications:
Jugalbandi does not require training an AI model
It's important to know that Jugalbandi doesn’t involve developing an AI model from scratch. Rather, it leverages pre-trained models such as GPT-4, Llama, Phi by integrating them into various applications. The accuracy offered by Jugalbandi is not due to the fine tuning of these models either, rather it relies on an approach called Retrieval Augmented Generation (RAG) that removes the need of training an AI model from scratch or fine tuning a model.
Jugalbandi application is not limited to government schemes
While Jugalbandi’s first application was a bot built for seeking information on government schemes, its utility extends well beyond this realm. Its conversational and multilingual interaction capabilities make it suitable for diverse sectors.
Jugalbandi is not confined to chatbot functionality on WhatsApp and Telegram
Jugalbandi is platform-agnostic, capable of integrating with a wide array of digital platforms, including mobile apps, web portals, and other messaging services. This flexibility ensures that Jugalbandi can meet users on their preferred platforms, providing consistent and accessible support across multiple channels.
Jugalbandi is not a simple plug and play API
Jugalbandi implementation goes beyond simple plug-and-play APIs that require zero to minimal coding for integration. While it offers robust functionalities and flexibility, integrating Jugalbandi into applications often involves tailoring and customization for specific usecases.
Jugalbandi code is free to use
If you wish to deploy Jugalbandi for your use case, the code is available freely on GitHub. One thing to note, however, is that all the components of the Jugalbandi stack are not entirely free of any cost. While many elements are open-source or low-cost, certain components like GPT models and cloud services may incur additional charges to be borne by the entrepreneurs building out their instance of Jugalbandi.