Featured · Startup
TasteCast
Cofounder + Technical Builder
TasteCast is an AI-powered demand forecasting and inventory optimization platform for small and independent restaurants. It helps restaurants predict daily item demand, plan inventory, reduce food waste, avoid stockouts, and make smarter prep or promotion decisions using data-driven forecasts.
TasteCast helps independent restaurants turn historical sales and transactional data into actionable forecasts and stocking recommendations. The product focuses on ease-of-use for small teams: daily demand predictions, clear actionable signals (BUY, OK, SURPLUS), and weekly advisories for prep and promotions.
Designed to reduce food waste, avoid costly stockouts, and simplify planning for kitchens with limited staff and storage.
- Built the core forecasting MVP using Python, pandas, and scikit-learn.
- Implemented CSV data ingestion and feature engineering pipelines to clean and transform raw restaurant data.
- Built demand prediction models and inventory/stocking signals that output recommendations like BUY, OK, or SURPLUS and weekly advisories.
- Designed backend logic that translates forecast outputs into actionable recommendations for operators.
Tech stack & architecture (summary)
High level: TasteCast ingests restaurant CSV data, canonicalizes it, runs per-store×menu-item forecasting, and translates forecasts into ingredient-level demand and inventory advisories.
Core backend responsibilities
- Ingest CSV uploads (multi-file or an all-in-one CSV) via an API endpoint (e.g.
/api/ingest). - Auto-detect upload mode: sales-only, multi-file (sales/recipes/ingredients/inventory), or all-in-one rows.
- Normalize/split incoming data into internal tables:
sales_df_raw,recipe_df_raw,ingredient_df_raw,inventory_df_raw. - Canonicalize fields (date, store, menu_item, qty_sold) and run per-store×menu forecasting.
- Use recipes + ingredient metadata + inventory snapshots to derive ingredient demand, simulate inventory, and emit BUY/OK/SURPLUS advisories.
Project layout
tastecast/ ├─ api/ │ ├─ ingest/ # endpoint + file parsing + mode detection │ ├─ forecast/ # forecast output endpoints │ └─ ingredient-plan/ # advisories, planning ├─ core/ │ ├─ preprocessing/ # column mapping, dedupe, date parsing │ ├─ forecasting/ # per store×item forecast logic │ ├─ recipes/ # menu-item to ingredient expansion │ └─ inventory/ # inventory simulation and advisories ├─ uploads/ # sample CSVs for tests ├─ schemas/ # CSV specs / field defs └─ frontend/ # upload UI, dashboards
This architecture cleanly separates ingestion, preprocessing, forecasting, and planning, allowing graceful degradation if only sales are available.