Menu Intelligence for Platform Operations
Automated catalog cleanup for multi-restaurant platforms. Standardize cuisine tags, identify duplicate listings, and monitor data quality across every restaurant in your network.
Catalog quality at scale
Multi-restaurant platforms accumulate millions of menu items from thousands of POS systems, each with its own naming conventions. The same dish appears as "Chicken Tikka Masala" on one restaurant, "Tikka Masala (Chicken)" on another, and "चिकन टिक्का मसाला Family Pack" on a third. Multiply that across every item in every restaurant and the catalog fills with silent duplicates that nobody catches until a customer complains.
Cuisine tags aren't any better. Restaurants self-tag, and there's no enforcement. A shawarma place tags itself as "Continental." A ramen shop picks "Chinese." Aggregators who depend on these tags for search filtering and category pages end up serving wrong results, and manual correction across thousands of menus is a losing battle.
The downstream effects compound. Duplicates inflate the catalog, making price comparison unreliable. Inconsistent tags break cuisine-based search filters. Missing data (no descriptions, no dietary labels, placeholder pricing) degrades the customer experience. And every week, new restaurants onboard with the same problems. Manual cleanup doesn't scale. Rule-based scripts break on edge cases. General-purpose text matching fails on food-domain specifics like transliteration variants, promotional noise, and cross-lingual dish names.
Automated pipeline for catalog quality
Latimal provides a three-stage pipeline that's purpose-built for platform catalog operations. Classification assigns standardized cuisine and category labels. Deduplication identifies duplicate listings across naming variations and scripts. Health reports flag missing data, outlier prices, and quality issues. All three work through a single API, processing items in bulk.
The system handles the food-domain specifics that generic tools miss. Cross-lingual matching works across 100+ languages and scripts, so "Pad Thai", "ผัดไทย", and "Phad Thai Noodles" all match correctly. Promotional noise stripping sees through "🔥 MEGA DEAL 🔥 Pepperoni Pizza [Buy 2 Save 40%]" to recognize the base item. Dietary conflict detection catches a "Paneer Tikka" listed as non-vegetarian, or a "Chicken Caesar Salad" marked veg.
Cleaner catalogs feed better outcomes everywhere downstream. Search becomes more accurate because duplicates no longer dilute results. Price comparison works because you're comparing the actual same dish across restaurants. Category pages show the right restaurants. And customers stop seeing three slightly different listings for the same butter chicken.
Key capabilities
Cuisine classification
Auto-tag every item with standardized cuisine and category labels. Tonkotsu Ramen gets tagged as specifically Japanese. Pão de Queijo gets Brazilian. Classification works across 100+ languages and handles regional naming conventions that vary by city and platform.
Classify API docs→Deduplication
Identify duplicate listings across naming variations, transliterations, and promotional noise. "Murgh Makhani", "Butter Chicken", and "**BUY 1 GET 1** बटर चिकन [Serves 2]" resolve to the same dish. The match endpoint handles cross-script comparisons, spelling variants, and packaging differences without rules or regex.
Dedup API docs→Health reports
Flag catalog quality issues before they reach customers. Missing descriptions, outlier pricing, dietary conflicts (a "Veggie Burger" listed under non-veg), and incomplete modifier trees. Run reports per restaurant or across your entire catalog to catch data rot early.
Report API docs→Purpose-built for food
General-purpose text matching treats "Kadhai Chicken" and "Karahi Chiken" as unrelated strings. It can't tell that "Pollo alla Parmigiana" and "Chicken Parmesan" are the same dish in different languages. It tags "Bibimbap" as generically Asian when it's specifically Korean. And it has no concept of dietary attributes, so a "Classic Burger" and a "Veggie Burger" look like duplicates when they're fundamentally different products.
Latimal was built specifically for this domain. The matching system understands food-specific semantics: transliteration variants across scripts, regional naming conventions (the same curry has different names in Mumbai and Chennai), promotional packaging that obscures the actual item, and the critical difference between a true duplicate and a dietary variant. That's what makes the difference between a dedup run that saves your ops team 200 hours and one that creates 200 new problems.
See it in action
Try the classification and dedup tools with your own menu data. Paste a list of dish names and see how the system handles transliterations, promotional noise, and cross-lingual variants.