Semantic FAQ Retriever

I built this project as a lightweight semantic search backend for FAQ data using TF-IDF embeddings and cosine similarity.

Features

• Semantic FAQ search endpoint
• Embedding preview endpoint
• Domain-tagged FAQ dataset
• FastAPI docs for quick demo and validation

Run

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8002

Open after startup (local machine only): http://127.0.0.1:8002/docs

UI Demo (local machine only): http://127.0.0.1:8002/ui

For Recruiters

• Local API docs: http://127.0.0.1:8002/docs
• Live deployment URL: add after deployment
• Suggested screenshots:
  • docs_home.png
  • search_response.png

Sample request POST /search { "query": "webhook api integration", "top_k": 3 }

Sample response { "query": "webhook api integration", "results": [ { "id": 3, "question": "How to connect webhook to API?", "domain": "integrations", "score": 0.81 } ] }

Recruiter Demo Script

Use this exact flow in a live demo:

1. Open /ui
2. In search box use: webhook api integration
3. Keep top_k = 3
4. Click Run /search
5. Explain: "The model ranks FAQ items using TF-IDF vectors and cosine similarity."
6. In embedding box use: how embeddings help retrieval
7. Click Run /embed
8. Explain: "This endpoint returns vector dimensions and preview values for semantic pipelines."

What to highlight while presenting:

• API-first design with FastAPI contracts
• Explainable ranking approach suitable for baseline production use
• Easy path to scale into transformer embeddings and vector databases

Future Improvements

• Add persistent storage for larger FAQ collections
• Add query logging and analytics for relevance tuning
• Upgrade to a hybrid retrieval strategy for better ranking quality