Towa

We simulate human ad feedback at scale — replacing focus groups with AI agents that think, feel, and react like real consumers. Test 10,000 ads overnight, not over 10 weeks.

Comment

Inspiration

Every marketer has faced that moment — hovering over the “Publish” button, wondering whether their campaign will soar or flop.
We were inspired by one question: what if we could see audience reactions before posting an ad?
With the rise of generative AI and advanced video understanding, we saw an opportunity to rethink advertising feedback collection — shifting from reactive surveys to predictive simulations.

Our goal: use AI to predict emotional and behavioral reactions before a single campaign goes live.

What it does

Towa is an AI-driven campaign simulation platform that predicts how different audience segments will respond to video ads — before they’re published.

It runs hundreds of simulated audience interactions across virtual personas to visualize:

  • Sentiment peaks and dips over time
  • Engagement likelihoods by segment (e.g., Gen Z, Fintech professionals)
  • Actionable insights such as emotional tone, clarity, and resonance

Towa helps brands test their creative work in a safe, simulated environment — a “wind tunnel” for advertising.

How we built it

We built Towa during the Generative AI in Advertising Hackathon @ AdWeek NYC 2025 using:

  • Figma Make for interface design and motion prototyping
  • TwelveLabs API for video understanding and semantic cue extraction
  • ElevenLabs API for voice tone and emotion modeling
  • LLMs (GPT-5, Claude) for persona simulation and feedback generation

Each ad is analyzed to extract emotional, visual, and contextual patterns.
Then our simulation layer predicts how 25+ AI personas will react — generating synthetic, data-driven feedback summaries.

Challenges we ran into

  • Designing realistic yet explainable AI persona behaviors
  • Creating meaningful visualizations of emotion and sentiment timelines
  • Integrating multiple APIs (TwelveLabs, ElevenLabs, LLMs) within a short timeframe

Accomplishments that we're proud of

  • Built an end-to-end prototype combining video analysis, simulation, and UX visualization
  • Designed a human-centric feedback dashboard that mirrors how agencies evaluate campaigns
  • Demonstrated how AI can accelerate creative testing while maintaining high interpretability

What we learned

We learned that combining multimodal AI (video + speech + language) can simulate audience perception with surprising realism.
Even a small dataset, when modeled through simulated personas, can reveal why certain creatives succeed — not just whether they perform.

What's next for Towa

We plan to expand Towa into a campaign-level simulation engine, allowing marketers to:

  • Integrating Zapier, HubSpot, and CRM data pipelines to automatically refine AI personas based on real engagement, audience insights, and lead quality.
  • Connecting with ad platforms like TikTok Creator Marketplace— allowing Towa to run pre-launch simulations that predict engagement rates, emotion profiles, and ROI before the campaign goes live.
  • Generating new creative variants using generative AI, guided by simulated emotional response maps.

Our vision: a world where every campaign is tested in simulation before launch — saving brands time, money, and creative uncertainty.

Built With

  • anthropic
  • elevenlabs
  • nextjs
  • python
  • supabase
  • twelvelabs
Share this project:

Updates