Repository navigation
Solution_Architecture_Report
GitHub Action edited this page Feb 11, 2026
·
3 revisions
The project follows Domain-Driven Design (DDD) and Onion Architecture to ensure modularity and maintainability.
Important
Map the repository layers to the architecture.
- Root Directory: Root
- Application Layer: Application -> Coordinators and Use Cases.
- Domain Layer: Domain -> Business Logic and Entities.
- Infrastructure Layer: Infrastructure -> Technical concerns and External Adapters.
- Data Access Layer: Data Access -> Persistence and Datasets.
- Entities: Autogen Models, Metrics, Schemas.
- Value Objects: Model Configurations, Evaluation Results.
- Domain Services: Registry Services, Model Promotion Logic.
- Jobs/Use Cases: TrainingJob, InferenceJob, HatchetInferenceJob, EvaluationJob, TuningJob.
- MCP Server: Autonomous tool coordinator (MCP Server Bootstrap).
- DTOs: Job Contexts, Request/Response schemas, MCP TextContent.
- External Services: MLflow (for tracking and registry), Hatchet (for orchestration), Kafka (optional messaging), LiteLLM (LLM provider), R2R (RAG engine).
- Adapters: Searchers, Splitters, Signers, Orchestration Workflows, MCP Tools (plan_mission, execute_code, run_tests, security_review, retrieve_context, index_code).
- Repositories: DatasetRepository, RegistryRepository.
- Datasets: Training data, Evaluation sets.
- Vector Database: R2R RAG (Knowledge Graph + Vector)
- LLM Provider: Autogen (Microsoft), LiteLLM (Gemini Pro)
- Embedding Model: Managed by R2R
Tip
Use Mermaid for dynamic diagrams.
graph TD
A[CLI/Job Trigger] --> B[Application Layer]
M[MCP Client] --> MC[MCP Server]
MC --> B
B --> C[Domain Layer]
C --> D[Model Implementations]
B --> E[Infrastructure Layer]
E --> F[MLflow / Registry]
E --> I[Hatchet Orchestration]
E --> R[R2R RAG / LiteLLM]
B --> G[Data Access Layer]
G --> H[Datasets]
I --> B
MC -- "dispatches" --> T[MCP Tools]
T -- "validates" --> S[Sandbox]
Template generated for Agentic workflows.