Search and Converse with the Policy Repository

 Background: 

The client is a public policy think tank organization of a government, which is tasked with catalysing economic development and fostering cooperative federalism among its states and regions. The main aim of the think tank is to conceptualize a road map with 7- and 15-year timelines and build strategies for the same with an implementation protocol involving the regional governments, which in turn would foster inclusive growth of all the regions/states and thus the country.

Project brief:

The client aims to have a knowledge portal (initially for 10 sectors, and eventually expand later), which contains all the policies, data, insights, and other documents from all the regions and states. This portal will serve as a knowledge resource for searching and understanding state-level and region-level policies and data for officials, bureaucrats, policymakers, etc. The client intends to set up a mechanism wherein people can search and converse with the portal (in natural language) for the above-mentioned purpose.

Solution:

To develop a Gen-AI-based tool that can provide answers through Natural language queries and help in understanding context, summarizing and extracting the required information, and providing semantic search, with non-reliance on specific keywords. The tool also would be capable of integrating and providing information from multiple documents. The tool also would have OCR (Optical Character Recognition) and language translation capabilities.

Healthark’s Role

Having previously worked in building the Knowledge Portal / Repository for the client, we are building and will be running/transferring the Gen-AI tool. Initially, we trialled multiple Foundation models (both open source and closed source) and selected the most optimal one. The models were pre-trained on the large corpus of diverse text data and fine-tuned on a specific policy dataset. A user-friendly query interface that allowed input in Natural Language (initiating with English, but further versions to contain regional / state languages as well). The tool was continuously evaluated against standards, and multiple iterations were followed based on metrics and test user feedback. The model is currently being scaled and deployed at the client environment and will be maintained and monitored by Healthark till the successful deployment of the final version, which would then include voice-based query and multi-language input/output capabilities.

Outcome:

Currently, in the deployment phase, the tool will function as a Subject Matter Expert (SME) for individuals to find and comprehend various regional and national policies, as well as their interconnectedness. This will further help in creating or modifying policies/regulations and implementing them.

Outcome

  • Identified 25 spoke towns for hybrid centers, through an in-depth analysis focusing on current market opportunity and future avenues for growth
  • Proposed a layout for the hybrid center that will operate on the hub-and-spoke model offering full range of services
Share on