AI Strategy & Architecture
This case study demonstrates building an Agentic RAG system and applying AI-first engineering to modernize a legacy platform.
What Is This Project?
PhoenixDX is a real-world case study in AI-first software engineering. The scenario: a legacy .NET monolith serving 40,000 users needs modernization into microservices — with only 5 engineers in 9 months, zero downtime, and payment frozen in Phase 1.
The Challenge
5 engineers, 9 months, 40K users, zero downtime. Traditional approach: impossible. AI-augmented: feasible.
The Innovation
Treat AI as infrastructure, not a tool. Build an Agentic RAG system that lets AI query 10 years of legacy system history.
The Result
2× effective capacity (5 engineers → 10 effective). 5 services extracted. AI generates 60–75% of migration code.
Case Study Components
Tech Stack Summary
| Component | Technology | Purpose |
|---|---|---|
| Vector Store | ChromaDB | Store 21K embedded artifacts for semantic search |
| Embedding Model | all-MiniLM-L6-v2 | Local embedding (384 dims), no API cost |
| MCP Server | Python + mcp SDK | Expose ChromaDB as tools for Copilot via stdio |
| REST API | FastAPI + Uvicorn | Same tools exposed as HTTP for web app |
| LLM (Dev) | GitHub Copilot | Free, built into VS Code, uses MCP tools |
| LLM (Web) | Gemini 2.5 Flash | Function calling for Agentic RAG on web |
| Web Framework | Next.js 16 + React 19 | Ask AI page with SSR |
| Project Data | SQLite (better-sqlite3) | Structured project docs, phases, risks, etc. |