Project idea · ungraded

Semantic Search for the Club Blog

Build a search box for a collection of blog posts that finds posts by meaning, not only exact words: "beginner python workshop" should also find "Intro to programming session". Compare keyword (TF-IDF) search with embedding-based search on the same queries.

You will use

Milestones

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

01Collect 50–200 posts (titles and text) into a JSON file.
02Build TF-IDF search and a small command-line interface that prints the top five results.
03Write 20 test queries with the posts you expect, and measure how often the right post is in the top three.
04Add character n-gram search for typos and compare the scores.
05Add embedding search with a sentence-embedding model and combine it with TF-IDF (for example by averaging ranks).

Stretch goals

Add filters (author, date, tag) alongside the text search.
Highlight the matching sentences in each result.
Expose the search as a FastAPI endpoint for the website.

Data

Use the Epoch blog posts you can access, or any set of articles you write or collect. Sentence-embedding models such as those in the sentence-transformers library download weights the first time they run.

Run it yourself

pip install scikit-learn numpy sentence-transformers
python search.py "beginner python workshop"