Mitibar

| In‑Scope | Out‑of‑Scope | |----------|--------------| | • Real‑time recommendation carousel on the screen (after login) • Personalization based on: - Past orders (if user is a returning member) - Selected preferences (sweet, bitter, strength) - Dietary filters (vegan, low‑sugar, non‑alcoholic) • AI model that scores every drink in the catalog (0‑100 relevance) • “Why this?” tooltip explaining key attributes (flavor profile, key spirit, ABV) • Ability to “Save for later” or “Add to order” directly from the carousel • Admin UI for setting business rules (e.g., push 20 % high‑margin drinks) | • Full‑menu AI voice assistant (future phase) • Dynamic pricing based on recommendation (out of scope for MVP) • Integration with external liquor‑brand recommendation APIs (use internal catalog only) | | • Backend API endpoint /recommendations?userId= returning up to 5 drinks • Model training pipeline (daily batch) • A/B test framework to compare with “no‑recommendation” control group | • Real‑time per‑user reinforcement learning (requires massive infra) |

: A built-in alternative to the standard Task Manager that allows users to view and terminate running system processes. mitibar

Drop a ❤️ if you’d rather celebrate Mitibar than another boring Sunday. - Set up feature store & Kafka topics

| Week | Milestones | |------|------------| | | - Finalize data schema (user prefs, drink taxonomy). - Set up feature store & Kafka topics. - Build admin “Boost” UI (React). | | 2 | - Develop recommendation service (FastAPI). - Implement content‑based embeddings. - Unit tests & contract tests for /recommendations . | | 3 | - Integrate collaborative filtering model (Spark batch). - Wire up API → model inference pipeline. - Front‑end carousel component & “Why this?” tooltip. | | 4 | - End‑to‑end QA (performance, fallback). - A/B test harness & dashboard. - Launch to 10 % pilot, monitor metrics. | | Post‑Launch (Week 5‑6) | - Expand to 50 % traffic. - Iterate on model (add seasonality features). - Gather feedback, refine UI. | - Implement content‑based embeddings

(মিতিবার) Noun. Etymology: From Miti (measure/date) + Bar (day).