The Kai Solution Server will act as an institutional memory for an organization using Konveyor, storing and analyzing solutions to migration problems encountered during application modernization. By leveraging the Model Context Protocol (MCP), this server will enable Kai to learn from previous solutions, improving the quality of generated code suggestions and increasing migration success rates over time.

The Solution Server is the first component in Kai that will provide value from gathering data on migration activities across the entire organization.  

When you think of Kai, consider that Kai intends to provide value in 2 broad areas:

• Local: Think of this as helping in the IDE with the application source code being migrated
• Identifying potential problems related to a goal of migrating to a new technology
• Extracting insights/context to aid in planning & fixing a problem
• Tools to aid in testing or improving a fix

• Organization-wide view: Think of this as an aggregated view of multiple developers across the org sharing information in the Hub
• Capture data related to resolving specific migration problems at scale for the entire org
• Develop algorithms/heuristics to look at migration activities across the entire organization and extra insights to improve future activities

The solution server is targeted at the “organization wide view”; its goals are:

• Provide a confidence level of how likely Kai is to provide a fix to a specific rule violation by tracking how well Kai has performed when faced with similar problems in the past by the organization
• Improve the effectiveness of rules where Kai is struggling to provide a fix. This will be achieved by improving the hint provided to an LLM to fix an issue by learning how the organization has solved similar issues.
• For example, LLMs are struggling to generate a solution to a problem that will be accepted by the Migrator.  The Solution Server will look to see if other Migrators encountered this issue in the past in a different application and if they ended up creating a fix. Ideally, this will be computed by capturing data on when a solution was proposed to a Migrator and modified.  The Solution server will seek to use the information of when other Migrators manually modified a solution and work with an LLM to extract hints from the solution to aid a future attempt in a different application.

See: Konveyor Kai Solution Server Approach

Github:

• Prototype solution server #746
• Confidence Level: Ability to identify issues that Kai shouldn’t attempt to fix #761
• Implement hint enrichment resource using successful Kai solutions #765
• Support auth in MCP solution server #769

What is the main user goal aka job to be done?

The Solution Server, as a component of Kai, aims to fulfill several main user goals or "jobs to be done" for different stakeholders within an organization engaged in application modernization:

• As a Developer (or Migrator), I would like to efficiently and successfully modernize applications by:

• • Leveraging my organization's cumulative knowledge to improve the quality of generated code suggestions and increase my migration success rates.
  • Gauging the difficulty of particular migration tasks before attempting them by accessing success ratios for specific rule IDs.
  • Receiving enhanced contextual hints for LLMs to improve code generation, drawing from descriptions of successful solutions, clarifications, and lessons learned from failed attempts within my organization.
  • Having Kai remember and learn each code generation that solves an issue, thereby reducing the number of calls to an LLM for similar or common generations and saving cost.
  • Accessing past fixes and insights from other migrators within my organization when LLMs are struggling to generate a solution for an issue.
• As an Organization, I would like to establish and leverage a collective institutional memory to optimize and improve our application modernization efforts by:

• • Having a central repository that stores and analyzes solutions to migration problems encountered across the entire organization.
  • Capturing data related to resolving specific migration problems at scale for the entire organization.
  • Developing algorithms and heuristics to look at migration activities across the entire organization and extract insights to improve future activities.
  • Improving the overall end-user experience for those using MTA for migrations.
  • Identifying common patterns differentiating successful and unsuccessful migrations across our applications (if data scale allows).
  • Flagging rules or migration tasks that consistently prove challenging, highlighting areas where automated approaches may be insufficient and require expert review.
  • Providing a confidence level for how likely Kai is to provide a fix to a specific rule violation, based on past performance.
  • Improving the effectiveness of rules by learning how our organization has solved similar issues, which helps in improving hints provided to an LLM.
• As a Migration Expert, I would like to be alerted to consistently challenging rules or migration tasks that exhibit low success rates, high developer modification rates, inconsistent solutions, or recurring rejected patterns, so I can provide targeted manual review or intervention where automated approaches fall short.

Content journey

https://spaces.redhat.com/display/MMSDOCS/Solution+Server