Project idea · ungraded

Course FAQ Chatbot

Build a chatbot that answers questions about this course — what a step covers, how Run and Submit differ, which step to do next — using RAG over the course's lesson text, and that cites the lessons it used. It should say "I don't know" when the answer is not in the material.

You will use

Milestones

Build the smallest working version first. Ticks are saved in this browser only.

01Split lesson texts into passages of a few paragraphs, each with an id and its lesson title.
02Index the passages with TF-IDF or a sentence-embedding model and test retrieval on 20 questions.
03Connect a chat model (a local model through Ollama or a hosted API) with a prompt that requires citations.
04Return structured answers (answer, cited ids, confidence) and reject ungrounded ones.
05Build a tiny evaluation set of question/expected-lesson pairs and report retrieval and answer accuracy.

Stretch goals

Keep conversation history so follow-up questions work.
Serve it with FastAPI and add a chat box to a web page.
Log questions the bot could not answer to find gaps in the lessons.

Data

Use the lesson articles of this course (each step's lesson) or notes from your own classes. For a local model without API keys, install Ollama and pull a small model.

Run it yourself

pip install langchain-core langchain-ollama scikit-learn
python faq_bot.py