How to Build an AI-Powered Fragrance Recommendation Mini‑Program with Zero Coding

Idea Sprout: Break Human‑Machine Barrier with a Mini‑Program Smart Recommender

We chose the “WeChat Mini‑Program ↔ Mini‑Store” connection to embed an AI‑driven fragrance recommender into a WeChat shop, leveraging the mini‑program’s ability to host pages, logic, and API calls while reusing the store’s product and payment features.

Getting Started: No Coding Experience Required

We used Vibe Coding (a “vibe‑coding” approach) with AI programming tools such as Cursor and Windsurf, which generate code from natural‑language prompts. The workflow pairs the AI editor with the WeChat Developer Tool for preview and debugging.

Tool Recommendations

Mini‑programs are not pure front‑end pages; they need page structure, component logic, API calls, and cloud‑function integration.

Project is in a “small‑scale” stage, so keep costs low; avoid expensive tools.

Choose tools that allow easy model switching (e.g., Cursor, Windsurf) to compare large‑model capabilities.

We also set up the WeChat Developer Tool for code preview and linked the same project folder between Cursor’s “Open Project” and the developer tool.

Defining the Product: Treat GPT/Claude as a Full‑Time Expert

We clarified the product concept “AI‑recommended fragrance” (named SCENTFLOW), drafted a system prompt that defines the model’s role, required context, and output format, and instructed the model to first think step‑by‑step before generating recommendations.

Running the Main Flow: AI‑Assisted Debugging

We iteratively ran the generated code, captured error messages, and fed them back to the AI for fixes. Common issues included navigation failures, sharing settings, and component misuse.

Key debugging steps:

Provide accurate error information to the AI.

Use code rollback features (restore checkpoint / revert) when needed.

Technical Challenges: Large‑Model Integration and Mini‑Program ↔ Mini‑Store Connection

We accessed large‑model APIs via WeChat cloud functions to protect API keys, prepared a structured product CSV, stored it in a cloud database, and designed system prompts to limit the model to recommend only items from that database.

We also followed the “Mini‑Program ↔ Mini‑Store” integration steps: open a mini‑store, upload products, obtain store and product IDs, and instruct the AI to embed the appropriate component on the result page.

UI/UX Refinement

We used AI to analyze reference images, generate design language, and translate it into WXSS/CSS. Visual assets were created with design tools or AI image generators, then imported into the mini‑program.

Final Steps: Publishing the Mini‑Program

After completing design and core flow, we registered the mini‑program, chose an appropriate subject (individual, sole proprietorship, or enterprise), completed the AI‑deep‑synthesis certification, submitted the code for review, and released it.

Within two weeks the SCENTFLOW mini‑program reached over 400 users and 200 sales, demonstrating that a non‑technical market professional can launch an AI‑powered e‑commerce product with minimal code.