Inspiration

Marketing performance often stalls when it comes to producing ad creatives. Bulk video creation can be labor-intensive and expensive, and top-level designers typically aren’t interested in repetitive ad iteration. Furthermore, ad reporting is time-consuming, with marketers juggling daily client requests for performance data and custom reporting. Drawing on our team’s background in marketing, we saw a chance for automation: an AI agent that slashes creative production costs and turnarounds, while simplifying performance analytics.

What It Does

Marketing Employee automates and streamlines marketing workflows:

  • Creative Generation: Transforms static images into fully produced video ads, complete with professionally synced voiceovers and subtitles.
  • Meta Ads Integration: Uploads the finished ads directly to your Meta Ads account
  • Data Retrieval and Analysis: Pulls real-time campaign metrics from Meta Ads and uses an LLM to provide immediate insights, which can be queried through voice commands.

By handling both creative production and ad performance in a single system, Marketing Employee helps marketers save time, cut costs, and manage more campaigns with fewer headaches.

How We Built It

  • Small Brain and Big Brain: We have a low-latency, 'small brain' AI agent that routes the user's requests. 'Big brain' covers specific use cases e.g. creative generation and has access to tools to achieve the use case.
  • AI-Driven Video Creation: We combined image-processing APIs and text-to-speech technology to create videos from simple image assets.
  • Voice-Subtitle Synchronization: We employed a speech-to-text alignment layer to add subtitles to videos.
  • Meta Ads API Integration: Multiple endpoints manage creative uploads and insights retrieval.
  • Conversational Interface: The user can chat with the report via voice or text to dive deeper into data and get optimization advice.

Challenges We Ran Into

  1. Subtitle Alignment: Getting subtitles to match perfectly with the voice track.
  2. LLM Hallucinations: Preventing the AI from fabricating performance data. We had to use the highest model o3-high to do this.
  3. Complex Meta APIs: Handling multiple Meta Marketing APIs and unifying them in a seamless user flow.

Accomplishments That We're Proud Of

  • Full Creative Automation: We have mutiplied creative output without relying on a creative agency or designer.
  • Live Conversational Insights: Marketers now get immediate, spoken feedback on campaign metrics—no more endless spreadsheets or hunting for data.

What We Learned

  • User-Centric Design: Even a powerful AI needs a user-friendly workflow that integrates seamlessly into a marketer’s day-to-day tasks.
  • Value of Guardrails: Fine-tuning the AI and building in constraints on data retrieval ensures reliability and maintains user trust.
  • Importance of Automation: When routine creative and reporting tasks are automated, marketers can focus on high-level strategy and creative direction.

What's Next for Marketing Employee

  • Post ads directly to Meta Ads
  • Pull ad-level data to inform which new ads should be created e.g. video view-through rate, conversions
  • Sophisticated video editor to make specific changes
  • Handle multiple videos in one go
  • Fix subtitle syncing
  • Other platforms e.g. TikTok, YouTube

Built With

  • decorator
  • elevenlabs
  • fal
  • fal-client
  • flask
  • groq
  • hypercorn
  • imageio
  • imageio-ffmpeg
  • llama
  • moviepy
  • numpy
  • openai
  • proglog
  • python
  • python-dotenv
  • python-telegram-bot
  • quart
  • requests
  • streamlit
Share this project:

Updates