Content Haus

Inspiration

With the rise of generative AI (Gen AI), the way we interact with information and social media is rapidly transforming. Gen AI can analyse vast amounts of data to produce entirely new content, from realistic images to captivating music. The societal impact is undeniable, with AI-powered tools democratizing creative expression and fueling innovation across various industries. However, this abundance of content also creates challenges for content creators struggling to stand out and gain a foothold on their social media presence.

As Tiktok users want more unique content and even more content creators flood the market, the need to elevate both creation efficiency and content quality becomes paramount. Just like user-friendly editing software empowers beginners, ContentHaus aims to be a one-stop shop for content creators. By leveraging Gen AI, it aims to bridge the gap between creators' visions and their final product, by providing storyboarding suggestions, and video editing.

What it does

Tech Stack

Front-end -React
- AntDesign -Back-end -Python FastAPI
- Docker
- Docker Compose
Cloud Services used under the Google Cloud Platform
- Cloud Run -Cloud Build
- Vertex AI
- Cloud Storage
- Cloud SQL
- Firebase

How we built it

Automated Discovery of Tiktoks

Content creators can find Tiktoks easily based on the criteria that they have chosen which may include: Hashtags, Current Trends or even Tiktoks from other fellow creators, without having to scroll through Tiktoks tediously on their own. Our discovery engine does the work for them gathering matching Tiktoks, which are used in our next feature.

Personalised AI-Generated Recommendations

Making use of the aforementioned collected Tiktoks, content creators can benchmark their own Tiktoks, regardless of whether it has already been published or in their drafts, against matching Tiktoks and the AI can generate personalised recommendations for them. The AI is not only able to provide general Tiktok improvements but also improvements specific to the criteria that they previously specified, ensuring that their Tiktok is still relevant to their original idea

Challenges we ran into

Unfamiliar with APIs

For us, this is the first time where we are working with Tiktok APIs, so there was some initial unfamiliarity with TikTok's various APIs and app registration interfaces. Additionally, the TikTok APIs were difficult to gain access to. In order to have proper access to the developer APIs, we had to gain approval from TikTok.

We therefore had to utilise open-source solutions which required quite a lot of tuning on our end to fit our purpose.

Processing Videos

We had difficulty working with some of the LLM models as it was not able to give us the desired output for our recommendations. Hence, we had to experiment with a few, ultimately deciding to settle for Google’s Gemini, which seemed to work the best for our use case - video processing.

Parsing the results was also challenging as oftentimes they came back in different formats. We had to fine-tune the prompts and regex used to ensure the output came out in a way that can be easily consumed by our users.

Scalability of Videos

Given our lack of funds, our application is unable to process longer videos fast. Our LLM instances take a long time for each video. Given the funding for stronger models, we would be able to utilise them to process our longer videos.

To tackle this, we stored the storyboards generated for each video to minimise the cost.

Lack of Domain Knowledge

Our team did not comprise of anybody who was familiar with creating content on TikTok. As such, we approached it from what we think is useful in our own eyes. In hindsight, it would have been helpful if we spent more time researching what TikTok content creators appreciate.

What we learned

Frameworks

For most of us, we are not familiar with FastApi, we wanted to take the chance to learn it, hence we decided to use it for this hackathon to learn more.

Domain Specific

The team also faced lack of knowledge of:

Not familiar with video editing
Not tiktok content creators
Not familiar with FastAPI

What's next for Content Haus

Currently, Content Haus sits as a separate application for creators to use. We envisioned it to be a feature integrated with TikTok, and as such, in the future, we hope for it to be implemented as part of the application, where content creators can dynamically generate their storyboards within the application and edit their videos, all within a few clicks of each other.