This repository contains a Retrieval-Augmented Generation (RAG) + few-shot prompt solution for automatically answering LINE Official Account Advanced certification exam questions via a Gradio UI.
-
Create & activate an Anaconda environment
conda create -n scrapeline-env python=3.11 -y conda activate scrapeline-env
-
Install Python dependencies
pip install -r requirements.txt
-
Install Playwright browser binaries
playwright install
-
Add Google Vision credentials
Save your Google Cloud Vision API JSON key file as vision-api.json in the project root.
The notebook uses a MODEL constant—make sure it matches the model you run below. I used qwen3:14b-q4_K_M:
bash ollama run qwen3:14b-q4_K_M
If you want to use a different Ollama model, replace
qwen3:14b-q4_K_Mabove and update
MODEL = "<your-model-name>"in the notebook.
You’ll need your LINE Business ID and saved cookies to pull down the official exam content:
-
Run and login
python save_cookies.py
-
Scrape each course’s questions
python scrape_shiken.py --course basic python scrape_shiken.py --course advanced python scrape_manabu.py
RAG alone wasn't enough to pass the advance certification, so I injected a small set of questions and answers I knew (plus a brief explanation):
-
Create a file called
few_shots.pynext to your main script:# few_shots.py SYSTEM_PROMPT = """\ You are an expert on the LINE Official Account Advanced certification exam. Your task is to read each “question + 4 answer choices” and use ONLY the materials retrieved from the vector database to produce the correct answer. Output rules: 1. If the question says “multiple answers,” select **all** correct choices. Otherwise select **exactly one**. 2. Do not output just “A/B/C/D”; list the full text of each chosen option. 3. Include a concise reason (≤ 50 characters). 4. Strict format: Answer: - <correct choice 1> - <correct choice 2> ← only one line if single-choice Reason: <very brief justification> """ FEW_SHOTS = [ ( "…sample question text 1…", """Answer: - …correct option full text… Reason: …short justification…""" ), # …add 3–8 more high-value examples… ]
After completing all of the above steps:
jupyter lab Open notebook.ipynb.
Run all cells to build the vector DB and set up the chain.
In the final cell, execute demo.launch() (or run python app.py) to start the Gradio interface.
Navigate to http://0.0.0.0:7860, paste your exam question + answer choices, and click 解答する.