SparkAI

• 

  SparkAI

Inspiration

The influencer marketing industry is on fire, projected to reach a staggering $24 billion by 2024 (1). This explosive growth is fueled by the increasing adoption of AI, with a whopping 63% of marketers planning to integrate it into their campaigns (1). It's clear that influencer marketing is a powerful tool, with 85% of respondents acknowledging its effectiveness, a significant rise from previous years (1).

Digging Deeper: The Creator's Perspective

But what about the creators themselves, the lifeblood of influencer marketing? We delved into a 2022 survey of 9,500 creators (2). The results revealed a dedicated bunch, with 36% spending a significant 1-5 hours per week crafting content, and a remarkable 5% pouring in a staggering 40+ hours (2). 43% of creators want to bring in outside help , including hiring freelancers , virtual assistants , consultants, team members, social media managers, graphic designers, photographers, and videographers. (3). Interestingly, 62% of niche creators believe specialization is the key to engagement and reach, highlighting the importance of targeting specific audiences (2)..

The Actual Problem: Creator Pain Points

To gain a deeper understanding of creator challenges, our team conducted face-to-face interviews with LinkedIn Top Voice, Instagram Influencers, and many more. Here's what we discovered:

1. Growing Engagement: Due to the high amount of time investment needed to make niche content, influencers suffer to get engagement on their post.
2. Too much Information with ambiguous prompts in LLM's: While creators appreciate LLM's ability to generate ideas with a high number of tokens, they find it frustrating to always be specific about what to create. Extensive back-and-forth discussions are needed before content is finalized.
3. Tool Overload and Disruption: The influx of various AI tools on the market is causing disruption in creator workflows. The need for a unified platform that seamlessly integrates text and image generation, referencing, and content posting emerged as a critical need.
4. Demo First, Execution Next: Content Creators want to see how it will look on a platform and not on a chatbot.

These insights sparked the inspiration for our solution. Recognizing the industry's growth, the creator's dedication to quality content, and the challenges they face, we aim to bridge the gap our platform Spark AI.

What it does

Our platform is a one-stop spot for creators, empowering them with Gemini-AI-powered tools that listen to their unheard words and cater to make content for niche audiences while streamlining the entire content creation process.

Spark AI takes a single prompt and provides the user with niche content tailored to their needs, a reference to help content creators build trust with their audience, an image to bring the content to life, and a feature to post it directly to the medium blog post. The Gemini API is capable of performing most of these tasks.

How we built it

We started by developing a flow diagram, backend/frontend architecture, and designing how the user would like to interact within our desired platform from start to end.
Here are the steps we did:

1. Build a Product Requirement Document: Here we drew all the research that we did before starting the project. Our PRD helped us to understand the Why, What, and How of the project.
2. Exploring Gemini APIs in Google AI studio: We started playing with Gemini API to see how we could structure an ambiguous statement into a niche statement. For that, we first tested a few prompts that help provide structure to ambiguous statements. We then leveraged this structured output from Gemini as input for another Gemini API to build content. This helped us generate content that has niche content text, and references and provides customer value like Actionable Steps, Analysis of content, Aspirational goals, and anthropological reasoning for the content. This structured output is then used to generate desired content, including text and images.
3. Exploring low-code tools: We aimed to implement the entire workflow with low-code tools so we used FlutterFlow for the front end, Buildship for the backend API builder, Prompt Refinements and adding responsible AI in Google AI studio, and explored ways to connect Text to Image models with these low code tools. For Text to Text, we employed Gemini AI, and for Text-to-image, Stability AI was utilized since VertexAI did not allow us to connect at Buildship due to an Oauth2.0 error 401.

4. Integrating the low code tools: As SparkAI followed the decoupled architecture, the frontend (FlutterFlow) and the backend (Buildship) communicated through APIs, allowing for flexibility and scalability in the system. This design choice enabled seamless integration of new features and updates without disrupting the overall functionality of the platform.

5. Testing with users: We shared our tool with content creators to test it for their use case, gather feedback, and integrate it into our front end.

Challenges we ran into

Through the hackathon, we ran into multiple challenges. Here are some key challenges that we faced as a team using Gemini & Vertex API

Here are the challenges that we came across frequently:

1. Adhering to the rules: Staying with no code and low code tools to build applications. This was a major limitation as our team was experienced with coding and had to go through various no-code/low-code platforms and finally pick one to build our desired platform.
2. Integrating APIs: Due to a variety of features we had to explore different APIs and find the most suitable one for our needs. After careful consideration, we decided to integrate the API that offers the best combination of functionality, reliability, and ease of use. This will allow us to seamlessly incorporate the desired features into our system and provide a seamless experience for our users.
3. Text to Image API authentication error We got stuck in passing the text to Image API authentication. The documentationhttps://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/api-errors has not mentioned this 401 error and ways to handle it. We tried enabling the vertex ai API and adding credentials with Oauth 2.0, enabling billing for all the APIs, setting up the Gcloud secret token, and then passing it to Buildship but still, we got this same 401 error. { "error": { "code": 401, "message": "Request is missing required authentication credential. Expected OAuth 2 access token, login cookie or other valid authentication credential. See https://developers.google.com/identity/sign-in/web/devconsole-project.", "status": "UNAUTHENTICATED", "details": [ { "@type": "type.googleapis.com/google.rpc.ErrorInfo", "reason": "CREDENTIALS_MISSING", "domain": "googleapis.com", "metadata": { "method": "google.cloud.aiplatform.v1.PredictionService.Predict", "service": "aiplatform.googleapis.com" } } ] } }
4. Hackathon Team's Collaboration: As our team came from different parts of the world it had difficulties collaborating. Our team was online almost 24/7 with different time zones. There could be dedicated individuals allocated to a few teams that can help teams get on track with timings.

Despite the challenges of distance and time zones, our team made a concerted effort to enhance communication through regular check-ins and updates. Utilizing messaging platforms and video calls, we ensured that everyone was on the same page regarding tasks, progress, and any roadblocks encountered. This proactive approach helped streamline our workflow and foster a stronger sense of unity among team members, ultimately leading to more effective collaboration and problem-solving.

Challenges Faced:

Accomplishments that we're proud of

As a team, we are proud of our ability to adapt, problem-solve, bring diverse opinions to foster creative, and collaborate effectively under challenging circumstances. Our platform stands as a testament to our hard work, creativity, and determination to address the needs of content creators in the ever-evolving landscape of influencer marketing.

1. Thorough Research: Heard actual needs of the user and acted upon it.
2. Building A PRD: From a vague discussion to thorough background research, analyze current competitors and build a PRD before building.
3. Agile Framework: Regular team meetings, multiple iterations, and finding ways to move forward.
4. New Tools Adoption: Despite new to almost all the tools, our team successfully embraced no code, low code, and API requirements for application development.
5. Finding ways to integrate as many features as possible: Efficiently integrated the chosen API into our platform.
6. Quick Fixing Our Way: Overcame authentication hurdles with Text to Image API by using the stability AI available in the Buildship for now.
7. Online 24/7: Improved team communication despite geographical challenges.

What we learned

This past few weeks have been nothing short of an adventure. Our team pivoted back and forth, learned various technologies, and came up with an end product. Learnings:

1. Flexibility in tool selection enhances adaptability.
2. Documentation first, implementation second: Thorough API evaluation is crucial for seamless integration.
3. Document technical findings along the way: Clear and comprehensive documentation is paramount for API authentication.
4. Virtual Environment could be a struggle: Effective communication strategies mitigate distance and time zone barriers.

What's next for IdeaSpark

1. Integrating Content Calendar: User can build the content in a single day and schedule the posting on a particular platform.
2. Maximizing Gemini 1.5 pro capabilities: User can input images, videos, and audio, and generate content in various ways.
3. Leveraging other Models Vertex AI has some really cool models in the model garden and making a full use case around the same.
4. Advanced Content Personalization: Develop features that allow users to personalize generated content further, such as incorporating user preferences, tone of voice, or branding elements. This level of customization can enhance user engagement and brand authenticity.
5. Community Engagement Features: Introduce community engagement tools that enable content creators to interact with their audience, gather feedback, and foster a sense of community around their content
6. Responsive UI: Due to the hackathon time constraint our User Interface has many loopholes. We desire to fix them and make them a more responsive web app.

You can find the PRD that we used to build this platform here: Google Docs Link.

As we received emails to have two videos for low code here are the links to a 3 min and 5 min video: SparkAI Explanation Video: Same as Demo Video Implementation of low code tool 5 mins: Low Code Tool

Built With

• ai
• buildship
• firebase
• flutterflow
• gemini