Gemini Business Card Merge Simplifies a Frustrating Duplicate Record Workflow.

Handling double-sided business cards becomes difficult when a bot fails to recognize that an English side belongs to the same person represented by the Chinese side. In a previous setup, the user sent the Chinese front and the bot saved it. When the user then sent the English back, the bot created a second record. The result was two half-empty entries for the same individual.

How users resolve name mismatches

Rather than building a thousand fragile regex rules to guess whether two names are identical, I let the user make that distinction. I implemented a “waiting room” state. When the first side arrived, the bot asked, “Is there a back side?” If the user confirmed, the bot retained that image in a user_states dictionary with a five-minute timeout, reflecting the brief attention span of users.

The real advantage appears when both images are sent to Gemini together. Trying to merge “Wang Daming” and “David Wang” through code leads to frustration. Providing both images in one prompt and allowing the large language model to handle the semantic merge offers the only rational solution.

Prompt logic for merging card data

The prompt logic used to instruct Gemini to consolidate the information into one tidy JSON object was:

You are an expert OCR and data extraction agent. I am providing you with two images: the front and back of a single business card.

Your task:

1. Extract all professional information from both images.
2. Merge the data into a single unified record.
3. If a field (like Name or Company) appears in both languages, combine them into a single string (e.g., "Chinese Name / English Name").
4. Ensure the output strictly follows the provided JSON schema.
5. Ignore any background noise or irrelevant text.

Return only the final merged JSON.

Why Gemini 1.5 Flash was chosen

For the coding part, I selected gemini-1.5-flash for its speed, ensuring the user does not lose track of the wait time. I simply combine both image parts into the generate_content call.

def generate_json_from_two_images(
 front_img: PIL.Image.Image,
 back_img: PIL.Image.Image,
 prompt: str) -> object:
 model = GenerativeModel(
 "gemini-1.5-flash",
 generation_config={
 "response_mime_type": "application/json",
 "response_schema": NAMECARD_SCHEMA
 },
 )
 front_part = Part.from_data(
 data=pil_to_bytes(front_img), mime_type="image/jpeg")
 back_part = Part.from_data(
 data=pil_to_bytes(back_img), mime_type="image/jpeg")
 response = model.generate_content(
 [prompt, front_part, back_part],
 stream=False,
 labels={"client_id": "namecard"}
 )
 return response

Turning data cleaning into prompt engineering

This method converts a messy data-cleaning duty into a straightforward prompt engineering task. No more duplicate entries, no manual deletion—just a clean, consolidated record.