BART LLM Meeting (Nov 12 2024) - brightcone/bart GitHub Wiki

BART LLM Meeting Report

Date: November 12, 2024

Meeting Overview

The meeting focused on recent updates and requests for the BART LLM project, particularly concerning data access, API integrations, and a new system for password resets. The discussion covered a range of technical challenges, including on-prem server access, data handling, and security protocols for the AI-driven solution.

Key Discussion Points

1. RemedyForce Data Access

  • Objective: Acquire data from RemedyForce to train the base LLM model on BART's internal servers.
  • Progress: Server access has been established, and all necessary software has been installed on BART's on-prem servers.
  • Next Steps: Team will need a data dump of relevant information stored on these servers. Data processing and training will occur strictly within BART’s internal network.

2. OneLogin Password Reset Process

  • Objective: Improve the current password reset process using AI capabilities, focusing on enhanced security and efficiency.
  • Challenges:
    • Currently, users locked out of all devices have to contact the Help Desk, which verifies identity through video calls or supervisor approvals.
    • Implementing facial recognition as an alternative solution was discussed, with emphasis on verifying users even when they are unable to access email.
  • Suggested Solutions:
    • Utilizing the OneLogin API to streamline password resets and allow users to reset directly from an app.
    • Implementing text-based links sent to a registered device, allowing facial authentication for password resets.
  • Concerns:
    • Potential issues with outdated user photos and challenges for employees without mobile devices.
    • Team aims to capture most common use cases, while handling exceptions through Help Desk.

3. Server and Application Access

  • Objective: Facilitate internal access for testing by exposing applications and APIs within the network.
  • Progress: Current access via IP addresses is being considered for use in testing environments.
  • Next Steps: Establish a dummy domain if needed, or continue using IP access for internal testers.

4. Internal API Integration

  • Objective: Integrate with RemedyForce API to enable the LLM solution to create and manage tickets directly in RemedyForce.
  • Actions Required:
    • Approvals for API access are pending; internal requests have been initiated.
    • Assistance from Kathy Kelly was suggested to streamline access setup and API key creation.
    • Documentation Reference: RemedyForce REST API

5. Next Steps

  • For Data Access: Garrett will coordinate with Thomas to manage data transfer to the server for indexing and training.
  • For OneLogin API Integration: A meeting is scheduled to secure approval for OneLogin API access.
  • For RemedyForce API Setup: Reach out to Kathy Kelly to obtain access and facilitate API connectivity.

Summary

The team made progress in accessing the on-prem servers and preparing for data integration with the LLM. There was a consensus on implementing facial recognition for password reset with the understanding that edge cases would continue to rely on Help Desk support. Action items were assigned for data transfer, OneLogin API integration, and RemedyForce API configuration.