Skip to content

POC of a chat bot using own data (with fine tuning + retrieval augmented generation)

Notifications You must be signed in to change notification settings

nicolaivicol/chat-own-docs-ft-rag

Repository files navigation

Solution to GovTech challenge (Gigahack hackathon)

POC of a chat bot using own data (with fine tuning + retrieval augmented generation).
The data is the full collection of documents from the https://servicii.gov.md portal.
This is a solution for the GovTech Challenge hackathon: https://gigahack.md/.

Final presentation

Example

The challenge

Gov Challenge - Streamline and elevate the customer experience.

The Electronic Governance Agency invites participants to develop within the EVO government application and a web interface a module based on AI technologies that assists citizens in the process of obtaining information about public services. The module should include a conversational interface that enables dynamic interactions with users, allowing them to ask questions, make requests, and seek the necessary clarifications to receive logical answers based on the descriptions of public services available at: https://servicii.gov.md.

Evaluation Criteria:

  • Understanding User Needs and Requirements: Work closely with the Electronic Governance Agency and relevant stakeholders to understand the specific needs and requirements of citizens. Identify the types of public services most frequently inquired about and the potential questions citizens might have.
  • Creating a Conversational Flow: Design a conversational flow that guides users through their queries. Define different user intents (e.g., querying a service, asking for clarifications, requesting forms), and map them to appropriate responses.
  • User Training and Support: Provide clear instructions on how to use the application and the web interface.
  • Scalability and Future Enhancements: As usage grows, ensure that the system can handle increased load. Consider adding more features over time.

Data

The project will utilize the extensive dataset from the Public Service Portal, including detailed information about various government services, service steps, terms, associated fees, legal frameworks, and more. This dataset will serve as the foundation for generating accurate and contextually relevant responses to user queries. Government services are freely accessible via a HTTP REST interface with OpenAPI description available at this address: https://servicii.gov.md/swagger/index.html In short, to get the data for all services, one can:

  1. list service categories using GET https://servicii.gov.md/rsspa-api/catalog/categories
  2. then list public services for each category using GET https://servicii.gov.md/rsspa-api/catalog/categories/{categoryld}
  3. then get details for each service in the category using GET https://servicii.gov.md/rsspa-api/public-service/code/{code} A sample public service description can be accessed here: https://servicii.gov.md/rsspa-api/public-service/code/006000022

About

POC of a chat bot using own data (with fine tuning + retrieval augmented generation)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published