amazon-chatbot-experience

Senior ML Platform / Applied AI Engineer interview-prep portfolio, built around the MangaAssist case study (an Amazon-style anime → manga reading-order recommendation product) plus 30+ topic deep-dives that cover the full system-design and operations surface.

📖 Browse online

The whole repo is published as a searchable Jekyll site:

Once Cloudflare Pages is configured, the URL will land here. See "Deploying" below.

The site is built from this repo's markdown by Jekyll + the just-the-docs theme — every folder is a section, every .md file is a page, and search works across all of it.

🧭 Where to start

Goal	Read this
Learn the case study cold	`01-problem-statement.md` → `16-final-summary.md`
Drill on system design	HLD-Questions and LLD-Questions
Prep behavioural / leadership	Leadership-Narrative and POC-to-Production-War-Story
Brush up GenAI eval	Evaluation-Systems-GenAI
Cost / latency tradeoffs	Cost-Optimization-Offline-Testing
See the working code	`mangaassist_web/` (Next.js + FastAPI) and `streamlit_app/` (21-page prototype)

🗂️ Layout

Path	What's there
`01-.md` … `16-.md`	The MangaAssist case study, in narrative order
Topic folders (e.g. `LLMOps/`, `Evaluation-Systems-GenAI/`, `RAG-MCP-Integration/`)	Notes grouped by interview competency. Each folder has an auto-generated `index.md` with a summary + interview talking points + file table
`mangaassist_web/`	Production-shaped Next.js 14 + FastAPI app
`streamlit_app/`	Streamlit prototype with 21 experimental pages
`mangaassist_3d/`	WebGL atlas exploration
`experiments/`	Per-experiment hypothesis + results
`_config.yml`, `Gemfile`, `index.md`	Jekyll config for the GitHub Pages site
`scripts/gen_docs_indexes.py`	Regenerates each folder's `index.md` after you add new notes

🛠️ Local preview of the docs site

# one-time
gem install bundler
bundle install

# serve at http://127.0.0.1:4000
bundle exec jekyll serve

After adding new notes to a folder, refresh the per-folder file index:

python scripts/gen_docs_indexes.py

The generator only rewrites the block between  and  markers, so any hand-written summary or interview talking points above them are preserved.

🚀 Deploying

This private repo can't use GitHub Pages on the Free plan (Pages on private repos requires GitHub Pro). The deploy runs on Cloudflare Pages instead — free, supports private GitHub repos, native Jekyll buildpack.

One-time setup (Cloudflare Pages)

Go to https://dash.cloudflare.com → Workers & Pages → Create → Pages → Connect to Git.
Authorize the Cloudflare Pages GitHub App for the SrikanthVelpuri/amazon-chatbot-experience repo.
Build settings: - Framework preset: Jekyll - Build command: bundle exec jekyll build - Build output directory: _site - Environment variables: RUBY_VERSION = 3.2.2 (also pinned in .tool-versions)
Click Save and Deploy. First build takes ~2-3 min (cold bundler). Subsequent builds are ~30 s thanks to dependency caching.
The site lands at https://amazon-chatbot-experience.pages.dev. Add a custom domain under Custom domains if you want.

After that, every push to main rebuilds and redeploys automatically — no manual step.

Local preview without Ruby

python scripts/build_preview.py
python -m http.server -d _preview 8765
# open http://localhost:8765

This builds 6 representative pages styled to match the deployed Jekyll site — fast sanity-check tool, not a full replacement.

Why not GitHub Pages?

.github/workflows/pages.yml still works as a workflow_dispatch runbook for the day this repo goes public or you upgrade to Pro. Until then, pushes don't trigger it.

📜 License

Notes are personal interview prep — not legal, security, or financial advice. The MangaAssist case study is a fictional design exercise inspired by Amazon's catalog and chatbot patterns; nothing here represents Amazon.