RAG ASP.NET Basic

This guide covers project structure, environment setup, running Qdrant (Docker), starting the Web App, setting OpenAI secrets, ingesting sample data, and testing with the provided question list.

---

📁 Project Structure

repo-root/
├─ WebApp/
│  ├─ RagAppBasic.csproj
│  ├─ Program.cs
│  ├─ appsettings.json
│  ├─ Controllers/
│  │   ├─ IngestController.cs      # POST /ingest/upload
│  │   ├─ QueryController.cs       # POST /query
│  │   └─ HealthController.cs      # GET  /health/keys
│  ├─ Models/                     # DTOs & config
│  ├─ Services/
│  │   ├─ Embedding/              # OpenAIEmbeddingProvider / OllamaEmbeddingProvider
│  │   ├─ Llm/                    # OpenAIChatProvider / OllamaChatProvider
│  │   ├─ Vector/                 # Qdrant REST client
│  │   └─ Text/                   # PdfTextExtractor, TextChunker
│  └─ (optional) Tools/           # simple function-calling tools (demo)
├─ SampleData/
│  ├─ ... (13 files)
└─ rag_question_list_1-13.txt

---

🛠️ Environment & Docker (Qdrant)

Requirements:

• .NET 8 SDK
• Docker Desktop (or Podman)
• OpenAI API key (or use Ollama for local LLM)

Run Qdrant with Docker

Option A — Quick Docker Run:

docker run -d --name qdrant \
  -p 6333:6333 -p 6334:6334 \
  -v qdrant_storage:/qdrant/storage \
  qdrant/qdrant:latest

Option B — docker-compose.yml (recommended):

services:
  qdrant:
    image: qdrant/qdrant:latest
    container_name: qdrant
    ports:
      - "6333:6333"
      - "6334:6334"
    volumes:
      - qdrant_storage:/qdrant/storage
volumes:
  qdrant_storage:

Run:

docker compose up -d

Verify dashboard: http://localhost:6333/dashboard

---

🚀 Run the Web App

From the repository root:

dotnet restore
dotnet build
dotnet run --project ./WebApp/RagAppBasic.csproj --urls http://localhost:5000

Open Swagger UI: https://localhost:7249/swagger

If you omit --urls, .NET may choose a random port—check the console logs.

---

🔑 Set your OpenAI Secret

A) .NET User Secrets (recommended for local dev):

cd WebApp
dotnet user-secrets init
dotnet user-secrets set "Provider" "OpenAI"
dotnet user-secrets set "OpenAI:ApiKey" "sk-...your_key..."
# Optional model overrides:
dotnet user-secrets set "OpenAI:EmbedModel" "text-embedding-3-small"
dotnet user-secrets set "OpenAI:ChatModel"  "gpt-4o-mini"

B) Environment Variables:

macOS/Linux:

export PROVIDER=OpenAI
export OPENAI_API_KEY="sk-...your_key..."
export OPENAI_EMBED_MODEL="text-embedding-3-small"
export OPENAI_CHAT_MODEL="gpt-4o-mini"

Windows PowerShell:

$env:PROVIDER="OpenAI"
$env:OPENAI_API_KEY="sk-...your_key..."
$env:OPENAI_EMBED_MODEL="text-embedding-3-small"
$env:OPENAI_CHAT_MODEL="gpt-4o-mini"

Alternative: Ollama (no key required)

export PROVIDER=Ollama
export OLLAMA_BASE_URL="http://localhost:11434"
# ollama pull nomic-embed-text && ollama pull llama3

Health check: Open GET /health/keys in Swagger → ensure HasApiKey = true (when Provider = OpenAI).

---

📥 Ingest SampleData

Via Swagger:

1. Open POST /ingest/upload
2. Select multiple files from SampleData/
3. Click Execute
4. You should see: { "Ingested": <number_of_chunks>, "Files": 13 }

Via cURL:

curl -X POST http://localhost:7249/ingest/upload \
  -F "files=@SampleData/1. returns_policy.txt" \
  -F "files=@SampleData/2. returns_policy_variation.txt" \
  -F "files=@SampleData/3. shipping_policy.txt" \
  -F "files=@SampleData/4. shipping_policy_vi.txt" \
  -F "files=@SampleData/5. warranty_policy.txt" \
  -F "files=@SampleData/6. product_coffee_maker_faq.md" \
  -F "files=@SampleData/7. product_headphones_faq.md" \
  -F "files=@SampleData/8. hr_pto_and_holidays.txt" \
  -F "files=@SampleData/9. hr_remote_work_policy.md" \
  -F "files=@SampleData/10. tech_glossary_search.txt" \
  -F "files=@SampleData/11. troubleshooting_wifi_router.md" \
  -F "files=@SampleData/12. onboarding_checklist.md" \
  -F "files=@SampleData/13. security_best_practices.txt"

The app auto-detects embedding dimension and creates a Qdrant collection like docs_basic_1536 to avoid “dimension mismatch” when switching embedding models.

---

❓ Test with Question List

Open rag_question_list_1-13.txt and try the questions via Swagger or cURL.

Swagger: POST /query with:

{ "question": "How long do I have to return an item?", "topK": 5 }

cURL:

curl -X POST https://localhost:7249/query \
  -H "Content-Type: application/json" \
  -d '{ "question":"When will I receive my refund after a return?","topK":5 }'

More examples:

• What time do orders need to be placed to ship the same day?
• Do you offer expedited shipping to P.O. Boxes?
• How long is the warranty coverage?
• What is the battery life of H7 with ANC on?
• How often should I descale the Coffee Maker X100?
• Giao hàng tiêu chuẩn trong bao lâu?

---

⚙️ Quick Tuning (Optional)

You can tweak these via environment variables:

export CHUNK_SIZE=800
export CHUNK_OVERLAP=120
export TOP_K=5
# (if implemented) export MIN_SCORE=0.25     # cosine similarity threshold

• CHUNK_SIZE ~800–1200 works well for prose.
• CHUNK_OVERLAP ~10–20% of CHUNK_SIZE.
• TOP_K: 3–5 for simple Q&A, 5–8 for broader/multi-hop questions.

---

🧑‍💻 Troubleshooting

• Qdrant not reachable → check Docker (docker ps), and ports 6333/6334.
• 401/403 from OpenAI → verify OPENAI_API_KEY.
• Hallucination / off-topic → increase strictness in system prompt, raise MIN_SCORE, reduce TOP_K.
• Duplicate contexts → enable dedupe (hash normalized text).
• Switch providers (OpenAI ↔ Ollama) → app should pick a collection name that includes the embedding dimension to stay compatible.