How I Built My AI-Based Gift Recommendation System

AIGiftWhisper is a web app designed to provide personalized gift recommendations based on WhatsApp conversations. It analyzes conversations, extracts relevant gift ideas, and generates Amazon affiliate links to make the purchasing process easier for users. In this post, I’ll walk you through the development of AIGiftWhisper, the challenges I faced, and how the system generates personalized gift ideas.

What is AIGiftWhisper?

AIGiftWhisper is an AI-powered tool that takes WhatsApp conversation files uploaded by users and extracts personalized gift suggestions. The system uses the DeepSeek API to process the conversation and relies on prompt engineering to create meaningful gift ideas. It then generates Amazon affiliate links for those gifts, so users can easily purchase them.

How the System Works

Upload a WhatsApp Conversation
Users upload their WhatsApp conversation file, which is then sent to the DeepSeek API. This is where the magic happens.
Processing with DeepSeek
The conversation is processed using carefully crafted prompts to extract meaningful gift ideas. The output is returned in a JSON format.
Personalized Gift Ideas
The AI analyzes the conversation for key themes, preferences, and interests, and generates a list of potential gift ideas.
Generating Amazon Affiliate Links
Once the gift ideas are extracted, Amazon affiliate links are generated for each suggestion. This makes it easy for users to buy the recommended items directly from Amazon, with a small commission going to the website.

The Challenges I Faced

1. Prompt Engineering Issues

The core challenge of this project was fine-tuning the prompt used to interact with the DeepSeek API. Since the AI needs to interpret casual conversations, some of the messages can lead to inaccurate recommendations. For instance, a simple message like “ahahha” might be interpreted as an indicator for a “comic book” simply because the user laughed, which is not ideal. As a result, I had to spend a considerable amount of time tweaking the prompts to ensure more accurate results.

2. Long Processing Time

Another significant challenge was the time it takes to generate a response. Due to the length of the conversations (which can sometimes be quite long), the processing time increases. This is compounded by DeepSeek’s response time, which can result in delays when generating gift suggestions. While DeepSeek was chosen mainly for its low price, it doesn’t offer the same speed or efficiency as more advanced models.

3. Choosing the Right Model

Ideally, if cost were not a constraint, the system would use GPT-4, as its advanced capabilities would provide more accurate and quicker responses. However, due to the high token count of WhatsApp conversations and the cost considerations, DeepSeek was the most suitable option.

The User Experience

1. Tutorial for Exporting WhatsApp Chats

One of the challenges in providing a seamless user experience was explaining how users could export their WhatsApp conversations. Since not all users are familiar with exporting chats, I created a tutorial that walks users through the steps of exporting and uploading the chat to the website. This ensured that users could easily use the platform without any confusion.

2. UI and Design

The user interface was designed to be clean and intuitive, with a simple drag-and-drop feature for uploading files and clear instructions for the next steps. This design choice aimed to make the experience as smooth as possible for users.

Future Improvements

Although the system is functional, there are areas for improvement:

Refining the Prompt Engineering: Fine-tuning the prompts to minimize incorrect recommendations (like the “ahahha” example) is an ongoing process.
Faster Processing: Exploring more efficient API solutions to reduce the response time, especially for long conversations.
Enhanced Personalization: Adding more advanced features, like understanding user preferences in more depth (e.g., specific hobbies or interests), could improve the personalization of the gift recommendations.

Conclusion

Building AIGiftWhisper was an exciting and challenging journey. While there were technical hurdles, the result is a tool that can provide personalized gift ideas based on WhatsApp conversations. With continuous improvements, this system could become a powerful tool for anyone looking for gift inspiration.

If you’re interested in trying out AIGiftWhisper, visit aigiftwhisper.com to get started.

Articoli sul Blog AI

How I Built My AI-Based Gift Recommendation System