### Launch of AI Assistant in Mobile AppMagnit has introduced its proprietary AI assistant, named Mёdik (Magic), directly into the "Magnit: Promotions and Delivery" mobile app. Developed in-house by the company's technology team using open-source technologies and a third-party commercial large language model (LLM), the assistant enables users to select products based on specific criteria, such as meal types. It also supports querying order status and resolving issues without contacting customer support.Future enhancements will expand capabilities to identify maximum discounts on items, provide in-store navigation tips, assist at self-service checkouts, and recommend cosmetics and skincare products tailored to individual skin characteristics. Magnit positions this as the first AI assistant launched in the grocery retail sector's mobile applications.### Technical Foundation and Initial ImplementationThe AI leverages a hybrid approach: open-source frameworks for core functionality combined with a commercial LLM for advanced natural language processing. This setup allows real-time product matching from vast catalogs, drawing on structured data like attributes, prices, and availability. Current features focus on query-based recommendations, transforming vague user inputs—such as "ingredients for dinner"—into precise assortments, thereby streamlining the shopping discovery process.Integration occurs natively within the app, which already handles promotions, delivery, and loyalty programs, as evidenced by its core role in Magnit's multi-format retail operations. This embeds AI into daily user interactions without requiring separate tools.### Implications for Product Feeds in E-CommerceAI assistants like Mёdik directly influence product feeds by enabling dynamic filtering and personalization at query time. Traditional feeds rely on static rules or manual curation, but LLM-driven matching processes user intent against feed attributes—price, category, dietary needs—accelerating relevance without exhaustive pre-tagging. This reduces latency in feed updates, as real-time catalog changes propagate instantly to recommendations.For grocery e-commerce, where assortments exceed thousands of SKUs with perishable or promotional volatility, such systems minimize stale data exposure. The assistant's criteria-based selection hints at vector embeddings or semantic search over feeds, enhancing discoverability of long-tail items that rigid feeds overlook. If you're looking for help with your own **product feed**, check out this blog: [/blog/product_feed/](https://notpim.com/blog/product_feed/).### Elevating Catalog StandardizationCataloging in retail often suffers from inconsistent standards across suppliers, leading to fragmented data. Mёdik's deployment enforces implicit standardization: by querying across meal types or skin features, it demands uniform attributes in backend catalogs—nutrition profiles, ingredient lists, dermatological tags. Over time, this drives upstream improvements, as incomplete data yields poor recommendations, pressuring teams to align with emerging schemas.In e-commerce, where 70-80% of catalogs stem from diverse vendors, AI acts as a quality gate. Non-standard entries degrade LLM accuracy, fostering adoption of protocols like GS1 or custom ontologies. Magnit's in-house build suggests proprietary refinements to handle regional product nuances, setting a benchmark for scalable catalog hygiene.### Enhancing Card Quality and CompletenessProduct cards in grocery apps frequently lack depth—missing allergens, pairings, or substitutes—limiting conversion. Mёdik addresses this by inferring completeness from interactions: incomplete cards fail complex queries, revealing gaps for iterative enrichment. Future skincare recommendations, for instance, will necessitate attributes like pH levels or hypoallergenic flags, compelling fuller, context-aware cards.This shifts e-commerce from descriptive to predictive cards, where AI populates missing fields via inference (e.g., extrapolating meal suitability from ingredients). Result: higher user trust and reduced returns, as recommendations align with real needs. For content infrastructure, it automates enrichment workflows, prioritizing high-traffic items. Ensuring your **product descriptions** are top-notch can make all the difference. Read more: [/blog/how_to_create_sales-driving-product-descriptions-without-spending-a-fortune/](https://notpim.com/blog/how-to-create-sales-driving-product-descriptions-without-spending-a-fortune/).### Accelerating Assortment Deployment SpeedSpeed in outputting new assortments defines competitive e-commerce, especially in promo-heavy grocery. Manual onboarding—testing feeds, cards, promotions—spans days; AI collapses this to minutes. Mёdik's discount-hunting feature, slated for rollout, scans live feeds for optimal matches, enabling instant surfacing of flash sales or seasonal introductions without recrawling.No-code elements amplify this: open-source bases allow drag-and-drop prompt tuning and rule overlays, bypassing developer queues. Retailers can A/B test AI behaviors on subsets of assortment, deploying winners app-wide rapidly. In Magnit's case, tying AI to self-checkout and in-store guidance foreshadows omnichannel sync, where app learnings optimize physical layouts in real time.### No-Code AI and Content Automation SynergiesNo-code platforms paired with LLMs lower barriers to AI deployment, as seen in Mёdik's open-source foundation. Retail tech teams configure behaviors via visual interfaces—prompt chaining for queries, integration hooks for order APIs—without deep coding. This democratizes content processes: marketers define recommendation logic, ops handle support flows, accelerating iteration.For e-commerce infrastructure, it unlocks generative content at scale: auto-generating card descriptions, promo copy, or personalized bundles from feed data. Magnit's support resolution via AI exemplifies this, preempting tickets by synthesizing order history and policies. Hypothesis: as models mature, no-code will standardize AI across chains, compressing development cycles from months to weeks while maintaining custom edges. Managing your data for these tools is made easier with a tool like a **price list processing program** - check out this article: [/blog/price_list_processing_program/](https://notpim.com/blog/price_list_processing_program/).Retailer's.ru reported the launch, underscoring its pioneering status in grocery. VentureBeat covered related workforce AI innovations, highlighting broader platform potential. Managing your e-commerce operations often relies on the correct format of your data. Check out our in-depth guides on the **CSV and JSON formats**: [/blog/csv-format-how-to-structure-product-data-for-smooth-integration/](https://notpim.com/blog/csv-format-how-to-structure-product-data-for-smooth-integration/) or [/blog/json-format-how-one-store-turned-chaos-into-fast-synchronization/](https://notpim.com/blog/json-format-how-one-store-turned-chaos-into-fast-synchronization/)The launch of Magnit's AI assistant highlights a significant trend toward leveraging AI for product discovery and enhancing the consumer experience, especially as it relates to e-commerce in the grocery sector. This move signals a push for catalog standardization and richer product data to feed AI models. For platforms like NotPIM, this underscores the increasing importance of product information management in supporting sophisticated AI-driven functionalities. We see this development as a positive step towards smarter and more efficient e-commerce operations.