Concept
«Stolik» is a Telegram channel that provides a weekly AI digest of Moscow’s coffee shops and restaurants. Every Thursday, the bot gathers fresh news from the city’s establishments and publishes a curated selection: new menus, seasonal offerings, tastings, chef dinners, brunches, masterclasses, and local events. Users get a clear plan for the weekend in just two minutes—instead of spending half a day Googling and ending up at a familiar spot—and venues gain visitors without manual promotion; their news is automatically included in the digest.
Data Sources for Media
For «Stolik, ” the goal is not simply to compile content about food, but to understand the audience’s needs, gather their feedback, and uncover new weekend menus and events from Moscow’s cafes and restaurants. Therefore, the sources are divided into five groups.
1. Trends and Demand: Wordstat Yandex / Google Trends — searches for: «things to do on weekends Moscow, ” „new restaurants in Moscow, ” „coffee shops with laptops, ” „where to go with kids / a partner.“ TGStat / LiveDune — competitor engagement rate and formats (KudaGo, „Dai Otsusit, ” „Afisha Restaurants“).2. Venue Telegram Channels and Food Aggregators: Primary source. Restaurants themselves post announcements, photos of new dishes, and special offers here.3. Aggregators and Guides (for checking menus, addresses, events): 2GIS, Yandex Maps — menus, operating hours, new openings. Timepad (Food section) — events and master classes. Allcafe.ru, Restoclub.ru, Gastronom.ru — reviews, menus, restaurant news.4. Visual Content (images for posts): Pinterest — references for a unified aesthetic: „cozy cafe interior Moscow, ” „warm food photography.“ AI generation (Midjourney / Flux) — cover art for issues and decorative elements. Prompts: warm lighting, cozy cafe, natural textures, film grain. Photos from venue Telegram channels — always cite the source or obtain permission from the venue.5. Audience Pain Points and Feedback: Reviews on Yandex Maps, 2GIS, Zoon, Restoclub — checking for issues like seating, power outlets, company practices, and comments under food channels' posts.
Separation into Human and Machine Processing
Machine Processing
Data Collection: Parsing Telegram channels and aggregator websites (2GIS, Timepad), gathering posts and comments from social media via API, and extracting search query frequency from Wordstat. Analysis: LLM filtering for relevance, entity extraction (date, price, address, link), de-duplication, thematic clustering (branch, masterclass, tasting, new menu), CV photo selection avoiding «plastic» stock imagery, and sentiment analysis of reviews. Generation: YandexGPT / ChatGPT for post and description copywriting, prompts for Midjourney / Flux, and automated post assembly using templates. Distribution: Automated posting to Telegram, formatting, and UTM tagging.
Human Processing
Curation and Fact-Checking: verifying dates, prices, addresses, links, filtering duplicates, and removed events. Expert Review: selecting the top 5–7 items from 20–30, and tailoring the angle of presentation to specific segments (couple, company, freelancer, tourist). Visual Control: rejecting unsuccessful AI-generated images and approving cover art. Audience Engagement: responding to comments, conducting surveys, and gathering feedback.
Guidelines for Automated Services
Tone of Voice (for AI text generation):Friendly, calm, like advice from a friend. Be specific: prices, addresses, dates. Avoid robotic phrasing.Visuals (image generation):Style — «urban gastro-visuals»: warm tones, natural textures, atmospheric shots. Prohibitions: plastic stock photography, overexposure, visual clutter. Prompts: warm lighting, cozy cafe interior, natural textures, film grain.Content Structure (auto-generation):🍽️Hook Title: The audience’s pain point or situation («Where to go when you don’t want to go to a bar») or a series title. 🍽️Main Content: 5–7 listings—name, date, price, address, description, link. 🍽️CTA: «Save post, ” „Share with friends, ” „Book a table.“Ethics and Fact-Checking:Only verified data. It is forbidden to invent events, discounts, or „special offers.“ Canceled events are removed. AI-generated images are not presented as real photos of venues. Links lead to actual booking pages.Formats:Automatic assembly of a post from 5–7 photo cards. Auto-generation of the issue cover (date + theme).
