A practical, code-driven companion to mastering search & recommendation in the age of AI.
If you build, study, or just geek-out over digital storefronts, this repo gives you:
| What? | Why it matters |
|---|---|
| Production-ready code for search & recommendation pipelines | Clone → run → benchmark or deploy |
| A book-in-progress (open-source!) | Learn the maths and the PyTorch behind modern IR/rec-sys—chapter by chapter |
| An autonomous research agent | Let the bot scour the web, cluster findings, and hand you a cited report |
E-commerce IR has exploded—from BM25 to bi-encoders, from naïve top-N to neural bandits—yet solid, narrative-driven code examples are scattered across blogs and papers.
This project stitches them together in one place, backed by real notebooks, tested pipelines, and readable prose.
information-retrieval-e-commerce/
├─ search/ # Vector & hybrid search pipelines (BM25 → RAG)
├─ recommendations/ # Baseline → bandits → neural CTR models
├─ deep-research-hybrid/ # Submodule – autonomous research agent
│ └─ documentation/ # LaTeX chapters for the agent paper
├─ ecom-ir-book/ # Submodule – book chapters + Jupyter notebooks
│ └─ docs/ # Compiled book manuscript (coming soon)
├─ bandits/ # Submodule – book chapters on bandits optimization for search + Jupyter notebooks
│ └─ Bandits_Book/ # Draft manuscript plus notebooks
| Sub-repo | Elevator pitch |
|---|---|
DeepResearchHybrid |
AI agent that loops through Plan → Search → Analyze → Synthesize to produce fully-cited research reports—complete with embedding-based topic discovery (semantic clustering + auto-labeling) and HyDE query expansion. |
ecom-ir-book |
Notebook-first “book” that starts with an embedding-MLP baseline, marches through contextual bandits, and lands in neural UCB—each chapter equal parts intuition, maths, and PyTorch code. |
# 1. Clone with sub-modules
git clone --recursive https://github.com/VladPrytula/information-retrieval-e-commerce.git
cd information-retrieval-e-commerce
read :)Issues and PRs are welcome—especially examples, benchmarks, or corrections to the book chapters. Please read CONTRIBUTING.md before opening your first pull request.