ShopWithUs

Make Money On All The Products In Your Content
Affiliate Links For The Products You Mention
Transparent and Simple Earnings + Analytics
Zero cost to join, Powered by OpenAI, Earn up to 10% on every purchase

Inspiration

It is incredibly challenging for content creators to monetize their work. Many creators rely on platforms like Spotify, YouTube, and TikTok to distribute their content and monetize through ads or paid subscriptions. However, these platforms take a significant portion of the revenue generated from these sources. Creators can join affiliate programs but the administrative work of building the brand partnerships can be time-consuming and difficult. They have to finding relevant affiliate programs and brands, apply and wait for approval, and then they need the tools and systems in place to accurately track their earnings.

What it does

Our proposed solution is ShopWithUs, an AI-based tool that automatically generates affiliate links for products mentioned in audio or video content, such as podcasts or YouTube videos. The tool uses speech-to-text technology to convert audio to text and then uses AI to identify product references within the text. It then creates affiliate links for each product mentioned and provides the content creator with a custom, easy-to-remember URL that can be shared with the audience. The creator earns a cut of the revenue from any purchases made through the affiliate links, and the company behind the tool also earns a fee. The target market for this tool is primarily long-tail to medium-sized content creators who do not have a dedicated revenue department. The company aims to eventually expand to all major affiliate programs and even smaller brands.

How it works

The goal we set was to listen for new podcast episodes, identify any products mentioned, then generate affiliate links for those projects.

This has been accomplished by using a chron-job to check RSS feeds in our database. When a new episode is added to the RSS, we download the MP3 and process it using Assembly.ai. Once the episode has been processed, Assembly calls a webhook with the transcript in JSON format. We take that transcript and feed it to GPT in segments. Each call returns a list of all of the brands and products mentioned in that segment. These results are then compiled and are converted into affiliate links through the Amazon Associates API.

Finally, each podcast has a page allowing listeners to view these links and how they were mentioned. Interested users can navigate to the amazon listing and if they purchase it, our creators earn a commission.

Challenges we ran into

The most notable challenged we faced was the process of identifying relevant entities in our transcripts. Standard Named Entity Recognition tools were not meeting our expectations and were missing a great deal of relevant brands. We struggled with this, working to find other providers, other algorithms and datasets. Eventually we thought, "Why don't we see how GPT-3 handles this?"

With only a little bit of tweaking, GPT was outperforming existing NER tools. It was identifying every possible brand mentioned in our transcripts. We were even able to have it output the results in JSON format so we could ingest it more easily.

Accomplishments that we're proud of

We are proud of being able to find a solution to the challenge of accurately identifying entities in our transcripts. By using GPT-3, we were able to achieve better results than existing NER tools, which was a significant accomplishment. This highlights our ability to think creatively and find innovative solutions to challenging problems.

Additionally, being able to format the output in a usable JSON format was a key success as it allowed for seamless integration with our existing systems and processes. This improved efficiency and made the NER process much more manageable.

Overall, we are proud of our ability to overcome the challenges and find a solution that not only met but exceeded our expectations.

What we learned

The biggest lesson we learned was the versatility of todays machine learning models. The ability of GPT-3 to outperform existing Named Entity Recognition tools showed us the power of advanced language models in handling complex NLP tasks. This highlights the importance of staying up-to-date with the latest advancements in the field of AI and ML and not rely on traditional tools.