How dotnet/eshopsupport works
Project type and purpose
• eShopSupport is a full-stack, microservices-style .NET Aspire solution demonstrating generative AI use cases (text classification, sentiment analysis, summarization, chat bots, synthetic data).
• It combines ASP.NET Core backends, Blazor WebUIs, a Python inference service, an identity provider, and data-generation/ingestion tools.
High-level file-structure analysis
• root (solution file, global props, docs, seeddata)
• docs/Architecture.pptx – existing slide deck
• seeddata/ – C# DataGenerator project + sample JSON datasets
• test/ – end-to-end and evaluation tests
• src/ – main projects:
• AppHost – .NET Aspire host & API gateway, loads Python Uvicorn apps and local LLM plugins (Ollama)
• Backend – ASP.NET Core APIs (Catalog, Ticket, Assistant, TicketMessaging) and EF Core DbContext
• IdentityServer – OpenID Connect provider and sign-in UIs for authentication/authorization
• CustomerWebUI – Blazor client for customers
• StaffWebUI – Blazor client for staff with real-time ticket messaging
• PythonInference – Python/FastAPI service exposing embedding and classification endpoints
• DataIngestor – console app to ingest seed data into the DB
• Evaluator – .NET console to run evaluation tests on AI responses
• ServiceDefaults – shared HTTP/AI client defaults (ChatCompletion, PythonInference, Qdrant, Backend clients)
Key insights from README
• Architecture revolves around .NET Aspire’s AppHost, hosting multiple resource types (HTTP, Python, local models).
• Demonstrates integration of Python services and small local models via Ollama.
• Plans for cloud deployment, but current focus on local dev with Docker, Python, .NET tooling.
• AppHost (API Gateway & Orchestrator)
• IdentityServer (OAuth2/OIDC)
• Backend Services
• Catalog API
• Ticket API
• Assistant API
• TicketMessaging API (SignalR/Redis PubSub)
• Databases & Caches
• Relational DB (SQL Server/Postgres)
• Redis (real-time messaging cache)
• Vector store (Qdrant for semantic search)
• PythonInference Service (FastAPI via Uvicorn)
• LLM Engine (Ollama local model host)
• Clients (Frontends)
• CustomerWebUI (Blazor WASM/Server)
• StaffWebUI (Blazor)
• Supporting Tools
• DataIngestor
• Evaluator
• Tests (E2E, Evaluation)
• UIs → AppHost → appropriate API endpoints
• AppHost ↔ IdentityServer for authentication/authorization
• Backend Services ↔ AppHost (hosted as resources)
• Backend Services → Relational DB (CRUD via EF Core)
• TicketMessaging API ↔ Redis Pub/Sub ↔ StaffWebUI (SignalR)
• Assistant API → ChatCompletionService (OpenAI or Ollama plugin)
• Backend semantic-search services → PythonInference embeddings endpoint or Qdrant
• PythonInference ↔ AppHost (hosted under /python)
• DataIngestor → Relational DB (seed data)
• Evaluator → Assistant API + ChatCompletionCache (test folder)
c. Architectural patterns and principles:
• API Gateway pattern (AppHost centralizes hosting of .NET, Python, and LLM resources)
• Microservices (distinct projects for UI, API, inference, identity)
• Separation of concerns (UI vs API vs AI vs identity)
• Event-driven messaging (Redis/SignalR for real-time updates)
• Extensibility (PythonInference and Ollama are swappable)
d. Technologies & frameworks:
• .NET Aspire / ASP.NET Core / EF Core / Blazor
• IdentityServer (OpenID Connect)
• Python (FastAPI, Uvicorn)
• Ollama local LLM hosting
• Docker (local hosting), Redis, Qdrant
• OpenAI ChatCompletion client (via ServiceDefaults)