Getting Started

Getting Started

1. Requirements

• Python 3.12 recommended, with pip and internet for first-time embedding model download.
• CPU embeddings work; CUDA is optional and requires a compatible CUDA-enabled PyTorch build.
• Source code and requirements_rag_2026.txt from the main repository.
• The first real model run needs to download BAAI/bge-small-en-v1.5 unless already cached.

2. Windows PowerShell — copy/paste

Run commands from the folder containing the .py files:

py -3.12 -m venv .venv-rag
.\.venv-rag\Scripts\python.exe -m pip install --upgrade pip
.\.venv-rag\Scripts\python.exe -m pip install -r requirements_rag_2026.txt
.\.venv-rag\Scripts\python.exe -m pip check

# Check dependencies (no model download)
.\.venv-rag\Scripts\python.exe .\RAG_Service_Corrected_2026_Diagnostics.py --doctor

# Built-in unit tests (no model download, requires NumPy)
.\.venv-rag\Scripts\python.exe .\RAG_Service_Corrected_2026_Diagnostics.py --self-test

# Download model on first run and perform tenant-aware demo retrieval
.\.venv-rag\Scripts\python.exe .\RAG_Service_Corrected_2026_Diagnostics.py --demo

Use .venv-rag\Scripts\python.exe directly if activation is blocked by PowerShell execution policy.

3. Linux / macOS

python3.12 -m venv .venv-rag
. .venv-rag/bin/activate
python -m pip install -r requirements_rag_2026.txt
python RAG_Service_Corrected_2026_Diagnostics.py --self-test
python RAG_Service_Corrected_2026_Diagnostics.py --demo

4. Jupyter / VS Code

Open RAG_Service_Corrected_2026_Diagnostics.ipynb. Select the Python kernel with the required packages and execute top-to-bottom. The notebook's RUN_REAL_DEMO variable is False by default; switch it to True only if you want to load the real embedding model and run Qdrant queries. Notebook cells are designed to avoid CLI argparse conflicts with ipykernel_launcher.py.

5. Run local FastAPI

Set a real, private token for at least one tenant:

# Run locally in the same terminal; do NOT commit the generated value
$env:TENANT_A_API_KEY = (& .\.venv-rag\Scripts\python.exe -c "import secrets; print(secrets.token_urlsafe(32))")
.\.venv-rag\Scripts\python.exe .\RAG_Service_Corrected_2026_Diagnostics.py --api

For a separate terminal, set $env:TENANT_A_API_KEY to the same private value through a secure local mechanism; see API-Reference for the request example. The server binds to 127.0.0.1:8000.

6. GPU optional

.\.venv-rag\Scripts\python.exe -c "import torch; print(torch.__version__, torch.version.cuda, torch.cuda.is_available())"
$env:EMBED_DEVICE = 'cuda'  # Only if torch.cuda.is_available() is True

For constrained or unsupported GPUs use $env:EMBED_DEVICE = 'cpu'. Obtain platform-specific PyTorch installation commands from https://pytorch.org/get-started/locally/ rather than assuming a wheel matches your driver.

7. Expected example behavior

• Tenant A retrieves passages about example API pricing/rate limits.
• Tenant B retrieves only its own export/security passages.
• Repeating the identical request for tenant A in the same session and top_k can hit the in-process cache.
• A prompt containing a known injection pattern is rejected by the heuristic detector.
• The output contains passage excerpts, not LLM-authored prose.

Next: System-Architecture and Testing-and-Validation.