A мессенджер clothing store with photo try-on

The customer chats in мессенджер, the assistant finds products in your catalog, shows them on the customer's own photo, and takes the order — orders land in mongoDb and Microsoft SQL.

  • Selfie try-on answers the biggest objection
  • Products found by description, not SKU
  • Orders placed in the chat — no site, no cart
  • Fast replies: frequent searches cached in redis
RedisMongoDB ToolGoogle Sheets ToolWhatsApp Business CloudMongoDB Atlas Vector StoreEmbeddings OpenAI
Connect mongodbatlas and goScriptera

How it works

To move data from Microsoft SQL to Edit Image automatically, use a ready-made scenario — no manual exports.

  1. Starts when: WhatsApp message trigger
  2. Check: Check if message is button or image
  3. If yes: Route interactive vs image message
  4. If yes: Parse button and user data
  5. Check: Route VTO vs order button tap
  6. If yes: Set VTO context in Redis
  7. Then: Prompt user to upload selfie
  8. If no: Order orchestration AI agent
  9. Then: Send order confirmation
  10. If no: Get VTO context from Redis
  11. Check: Is VTO product ID stored in Redis?
  12. If yes: Extract image ID, waId, and product ID
  13. Then: Get product image from MongoDB
  14. Then: Download product image from Drive (VTO)
  15. Then: Resize product image to 1024px
  16. Then: Extract product image as base64
  17. Check: Merge product, selfie, and validation result
  18. Check: Validate merged selfie before try-on
  19. If yes: Build Gemini image generation API payload
  20. Then: Generate VTO with Gemini API
  21. Then: Convert Gemini response to image file
  22. Then: Upload VTO result to WhatsApp
  23. Then: Send VTO result to user
  24. Then: Delete VTO context from Redis
  25. If yes: Notify user VTO is processing
  26. If no: Ask user to resend valid photo
  27. Then: Get WhatsApp media URL
  28. Then: Download user selfie
  29. Then: Analyze user selfie with Gemini
  30. Check: Gemini: confirm exactly one person
  31. Then: Resize user selfie to 1024px
  32. Then: Extract user selfie as base64
  33. If no: Ask user to send correct photo
  34. If no: Validate incoming message
  35. Check: Pass valid messages, block invalid
  36. If yes: Get user session from Redis
  37. Then: Load session and append user message
  38. Then: AI shopping agent (BytezBot)
  39. Then: Detect JSON intent or plain text reply
  40. Then: Append AI reply to session history
  41. Then: Save session to Redis
  42. Check: Route JSON intent vs plain text reply
  43. If yes: Route product search vs recommend
  44. If yes: Build Redis cache key from search query
  45. Then: Check Redis cache
  46. Check: Is product cached in Redis?
  47. If yes: Parse cached product JSON array
  48. Check: Loop through products
  49. Then: Build product card message body
  50. Then: Convert product image base64 to binary
  51. Then: Upload product image to WhatsApp
  52. Then: Send product message with buttons
  53. If no: MongoDB Atlas vector search
  54. Then: Download product image from Drive
  55. Then: Convert product image to base64
  56. Then: Aggregate all product base64 images
  57. Check: Combine vector search results with images
  58. Then: Combine products with base64 images
  59. Then: Store products in cache
  60. Then: Unpack products from cache result
  61. If no: Send text response to user
  62. If no: Send validation error to user
  63. Then: GPT-5-nano (intent classifier)
  64. Then: GPT-4o (order agent)
  65. Then: Create order in MongoDB
  66. Then: Get product info from MongoDB
  67. Then: Log order to Google Sheets
  68. Then: OpenAI embeddings
  69. Then: Session memory for shopping agent
  70. Then: Session memory for order agent

You can launch this Microsoft SQL + Edit Image integration in Scriptera: describe the task in plain words — the scenario is built, launched and monitored for you.

Similar scenarios