24001 Enabling Conversational Analytics on Indian Policy Insights Projects using LLMs

Project Title: Enabling Conversational Analytics on Indian Policy Insights Projects using LLMs

Key Words: GeoAI, GenAI, NLP, ChatGPT, LLMs

Research Topics: GenAI, GeoAI

Mentors:

  • S V Subramanian, School of Public Health, Harvard University
  • Devika Kakkar, Center for Geographic Analysis, Harvard University

Project Description (IPI):

The India Policy Insights (IPI) project provides a comprehensive data platform that uses novel statistical techniques to measure the performance of population health and development indicators relevant to policy making in India. The data is extremely rich and presented at multiple geographic levels to provide insights. The project’s mission is to empower elected officials (National and State) to constructively engage with their constituents on individual and local health and well-being concerns. IPI resources are fully and openly accessible to the public for their use. Here is a more detailed description about the project and its data.

Tasks Description:1. Using open-source LLMs to enable Conversational Analytics on India Policy Insights
The intern/fellow will be closely working with CGA’s Data Science Manager to accomplish the following:

1.1. Understanding Data from Indian Policy Insights Project
Understand the scope and objectives of the IPI project. Getting familiar with the data and dashboards from the project at different geographical levels (Assembly Constituencies, Parliamentary Constituencies, Districts, and Villages) in India. Reviewing the dataset structure and contents, including the variables, units, and geographical classifications.

1.2. Building a Chatbot Using llama or other open-source LLM

Develop a functional chatbot powered by llama or other LLMs to support user queries in Natural Language (English and Hindi). The chatbot will integrate with existing dataset, dashboard and website, enabling conversational GenAI to answer user queries, perform tasks, and enhance user engagement. This would include the following:

  • Familiarization with ChatGPT Educational offering:
    • Review the functionalities and tools available within the ChatGPT Educational toolbox.
    • Understand the available model and AI capabilities that can be applied to IPI data
    • Explore building CustomGPTs for IPI projects using the educational offering.
  • LLMs/ Natural Language Processing (NLP):
    • Utilize LLMs capabilities for natural language understanding and generation.
    • Configure the chatbot to handle different types of conversations, including FAQs, instructions, and casual conversations.
    • Customize prompts to align with the specific use case of IPI project
  • Backend Integration:
    • Integrate llama or other LLM interfaces with the chatbot.
    • Set up API calls to send user queries to send and receive responses.
    • Implement error handling for cases such as API downtime or failed responses.
    • Add message rate limiting or throttling to prevent overuse or spamming.
  • Chatbot Frontend:
    • Design and implement a user-friendly chatbot interface (web-based or app-based).
    • Ensure the chatbot is performant, responsive and works on multiple devices (desktop, mobile).
    • Provide user options like text input, send button, and conversation history display.
    • Add indicators for typing and feedback on sent/received messages.
  • User Experience (UX):
    • Allow users to reset or clear conversations as needed.
    • Incorporate a feedback mechanism for users to rate the usefulness of responses.
    • Optionally add personalization features, such as remembering user preferences or previous conversations.
  • Data Management:
    • Ensure the chatbot logs conversations for further analysis (if required) while adhering to privacy and data protection guidelines.
    • Implement encryption for secure transmission of user data.
    • Provide options for user consent on data storage and usage.
  • Customization and Scalability:
    • Design the chatbot to be scalable, able to handle multiple users simultaneously without performance degradation.
    • Allow easy customization of the bot’s behavior, including adjusting its tone, knowledge base, or response style.
  • Testing and Validation:
    • Thoroughly test the chatbot functionality across various devices and browsers.
    • Validate the performance of the API integration and ensure seamless user interaction.
    • Test with a variety of conversation types to ensure accurate and relevant responses from LLMs.
  • Documentation and Deployment:
    • Provide clear documentation for setting up, configuring, and deploying the chatbot.
    • Deploy the chatbot on a live server or website with appropriate monitoring tools for usage statistics and error tracking.
    • Peer reviewed paper, presentation, workshops/tutorials on using LLMs on Geospatial dataset with IPI as a use-case

Deliverables:

  1. Fully functioning chatbot integrated with LLMs.
  2. Documentation covering setup, API integration, and customization.
  3. Test reports verifying performance and user interaction quality.
  4. A demo or deployed version of the chatbot on a website or application.
  5. Peer reviewed papers, presentations, workshops/tutorials related to the project.