FAQ System Architecture
A frequently-asked-questions bot answers a user by finding the stored question that best matches what they typed and returning its answer. This diagram traces the query "how do I reset password" through input processing, semantic matching against an FAQ database, and response delivery, then shows how user feedback flows back to improve the system. Hover over any box for details.
Interactive Demo
You can embed this MicroSim in your own page with the following iframe:
1 | |
Overview
The pipeline has five stages, color-coded in the diagram:
- User interface (purple) — the chat window where the user asks a question and later reads the answer.
- Input processing (blue) — text normalization (lowercasing, punctuation removal), synonym expansion, and embedding generation turn raw text into a numeric vector.
- Semantic matching (green) — similarity calculation, ranking, and a confidence threshold compare the query vector against stored FAQs.
- FAQ database (orange) — curated question-and-answer pairs stored as embeddings; the matching engine queries it and receives candidates.
- Response delivery (teal) — the top-ranked answer is returned (here with confidence 0.87), and thumbs-up/down feedback loops back to the database.
The dotted feedback arrow is what lets the system improve over time: answers that users mark unhelpful can be revised or re-ranked.
Lesson Plan
- Trace the flow: Have students follow the query from the chat window to the returned answer, naming each stage.
- Explain semantic matching: Why use embeddings and similarity instead of exact keyword matching?
- Discuss the threshold: What should the bot do when no FAQ scores above the confidence threshold?
- Analyze the feedback loop: How does thumbs-up/down feedback make the system better, and what could go wrong with it?