QScript Coding Assistant

Local DGX Spark setup for:

• vLLM with Qwen/Qwen3-Coder-30B-A3B-Instruct
• FastAPI backend
• React/Vite frontend

Install Environment

Create and populate the root .venv:

./install.sh

The installer uses Python 3.12, keeps the uv cache in the project, installs the vLLM/backend dependencies, and creates .env with DGX defaults.

Folder Structure

in the root directory create a spec folder with following contents
-spec (this in the root-dir)
--pdfs (CheatSheet.pdf)
--pdfs-txt (copied pdf text in *txt Files with Codierrichtlinien.txt)
--reports (specs as *.txt many shot examples)

Start Services

Start the services in this order.

1. vLLM Backend

From the project root:

chmod +x run-backend-qwen3-coder-30b.sh
./run-backend-qwen3-coder-30b.sh

vLLM runs on:

http://127.0.0.1:8042/v1

Check it with:

curl -sS http://127.0.0.1:8042/v1/models

2. FastAPI Backend

From the project root:

./run-api-dgx.sh

FastAPI runs on:

http://127.0.0.1:8080

Check it with:

curl -sS http://127.0.0.1:8080/health
curl -sS "http://127.0.0.1:8080/api/backend?provider_id=dgx_vllm_qwen"

3. React Frontend

From the project root:

./run-frontend.sh

React/Vite runs on:

http://127.0.0.1:5173

On the DGX network it may also be available at:

http://192.168.1.196:5173

Start Services With tmux

Use these commands to keep all three services running in the background:

tmux new-session -d -s qscript-vllm 'cd /data/Projects/qscript-coding-assistant && ./run-backend-qwen3-coder-30b.sh 2>&1 | tee /tmp/qscript-vllm.log'
tmux new-session -d -s qscript-api 'cd /data/Projects/qscript-coding-assistant && ./run-api-dgx.sh 2>&1 | tee /tmp/qscript-api.log'
tmux new-session -d -s qscript-frontend 'cd /data/Projects/qscript-coding-assistant && ./run-frontend.sh 2>&1 | tee /tmp/qscript-frontend.log'

List sessions:

tmux list-sessions

Stop services:

tmux kill-session -t qscript-vllm
tmux kill-session -t qscript-api
tmux kill-session -t qscript-frontend

View logs:

tail -n 120 /tmp/qscript-vllm.log
tail -n 120 /tmp/qscript-api.log
tail -n 120 /tmp/qscript-frontend.log

Important URLs

vLLM models:      http://127.0.0.1:8042/v1/models
FastAPI health:   http://127.0.0.1:8080/health
FastAPI backend:  http://127.0.0.1:8080/api/backend?provider_id=dgx_vllm_qwen
Frontend:         http://127.0.0.1:5173

Note: http://127.0.0.1:8080/v1/models is not a valid route. Port 8080 is FastAPI. The vLLM OpenAI-compatible API is on port 8042.