docs/PUBLISHING.md

Publishing & deployment guide

Step-by-step to get this on GitHub and (optionally) live on the web. Run these from the project root on your own machine (where git, python and npm work).

1. Add a screenshot (30 seconds)

The README shows docs/screenshot.png. To create it:

• Open docs/ui-preview.html in your browser, or run the app (see README),
• Take a screenshot and save it as docs/screenshot.png.

2. First commit

A fresh, clean history looks best on a portfolio repo. From the project root:

git add .
git commit -m "News search engine: TF-IDF retrieval, query expansion, FastAPI + React"

If git status shows leftover lock/config issues from the old repo, you can start a brand-new history:

rm -rf .git
git init -b main
git add .
git commit -m "News search engine: TF-IDF retrieval, query expansion, FastAPI + React"

Confirm the large files are ignored — git status should not list data/News_Category_Dataset_v3.json or any .rar/.pkl.

3. Create the GitHub repo and push

Using the GitHub CLI:

gh repo create news-search-engine --public --source=. --remote=origin --push

Or manually: create an empty repo on github.com, then:

git remote add origin https://github.com/<your-username>/news-search-engine.git
git push -u origin main

4. (Optional) Deploy a live demo

A live URL is the single most valuable thing to put on LinkedIn. The app is designed to deploy as one service (FastAPI serves the built React app).

Option A — Render (free tier, simplest)

1. cd frontend && npm install && npm run build && cd .. (creates frontend/dist). Commit frontend/dist for the simplest deploy, or run the build in Render's build command.
2. Push to GitHub.
3. On render.com: New → Web Service → connect the repo.
   • Build command: pip install -r requirements.txt && python scripts/build_index.py && cd frontend && npm install && npm run build
   • Start command: uvicorn api.main:app --host 0.0.0.0 --port $PORT --app-dir .
4. Render gives you a public https://news-search-engine.onrender.com URL.

The demo runs on the committed sample by default — light enough for free tiers. Skip --bert in deployment; the BERT model + PyTorch exceed free-tier memory.

Option B — Railway / Fly.io

Same idea: one web service, build the frontend, run uvicorn. Add a Procfile with:

web: uvicorn api.main:app --host 0.0.0.0 --port $PORT --app-dir .

5. Polish the repo page

• Add a short About description and topics on GitHub: information-retrieval, search-engine, tf-idf, nlp, fastapi, react.
• Pin the repo on your GitHub profile.
• Add the live demo URL to the repo's website field and to the README top.