Adsett

Transforming digital reactions into actionable campaign insights.

Comment

Inspiration

Upon the conclusion of the Hackathon kickoff, our team pondered on ideas that we could implement that would satisfy the conditions of the established tracks to success. However, every idea we found ourselves thinking the same thing; "how would we present this"? After all, it was universally agreed that the presentations would be the most critical determining factor to those judging our product. Eventually someone substituted the word advertising instead of presenting, leading to a slow but escalating cascade of ideas culminating in Adsett. What started off as a way to organize our pitch, and perhaps those of fellow hackers, transformed into a digital marketing tool helping consolidate cross-platform marketing assets and providing actionable insights for all advertisers.

What it does

Adsett allows users to organize marketing campaigns as typically there tend to be long-term, continuous campaigns, seasonal campaigns, and experimental campaigns. The user can upload photos of their marketing asset prototypes which immediately begins to receive initial insights. These insights use context about the campaign to complement and critique the marketing philosophy before the user deploys it on social media from the click of a button through Adsett.

Once deployed, our custom hybrid sentiment mode, named SynFusion analyzes public reactions such as likes and comments through numerous filters to extrapolate statistics such as resonance with the audience, hostility against the message or design, and any other detected criticism. We visualize these through easy to digest ring graphs with intuitive explanations explaining how the ratings were determined.
From here users can continue to optimize their marketing campaigns.

How we built it

AdSett was developed using the Next.js web development framework. For our front-end, we utilized TypeScript for performance and scalability and TailwindCSS to simplify coding a clean, responsive interface. The project follows a modular monorepo structure wherein our front-end is currently decoupled from the back-end for rapid iteration during the Hackathon.

On the back-end, we primarily utilized Python. For the digital sentiment hybrid model SynFusion, we utilized multiple NLP transformers pre-trained on relevant aspects such as controversy analysis and refined these separate models using logistic regression to create a more engaging analysis model. For AI-driven insights we used utilized FastAPI and Gemini and conditioned user-friendly outputs using LangChain, removing jargon and clutter for the initial insights and post-deployment analysis. For authentication we utilized Auth0.

Challenges we ran into

From a technical perspective a lack of data presented challenges when training the SynFusion sentiment hybrid model, leading to us generating human-reviewed synthetic digital engagement data to train the model. Authentication was challenging as we had not used Auth0 before and found it challenging to use with the multiple new technologies we were experimenting with already.

From a team perspective half of us were completely new to hackathons with the second half having only attended one. We were not accustomed to such a rapid development cycle and had to sacrifice much of our sleep time, leading to us believing what little sleep we did achieve was critically endangering the development of our project.

Accomplishments that we're proud of

We are proud of the development of the hybrid model in such a relatively short time due to it's accuracy. We are also proud on even developing an idea that we were not immediately rejecting and that we though would be a strong contender in the Hackathon. However, most importantly, we are proud that despite all the hurdles in our development life span we did not give up on the project. That despite things seeming impossible to complete, we continued in our effort to persevere.

What we learned

Throughout the development of Adsett, we learned how to design and scale architectures under critical time constraints and how to bridge front-end visual clarity with complex back-end intelligence. We deepened our understanding of data-driven UX design, prompt engineering for sentiment pipelines, and the importance of creating intuitive tools that balance analytics and learning. We also gained valuable experience in working as a team under strict time restraints.

What's next for Adsett

Moving forward, we plan to fully integrate real-time social media advertiser APIs, enabling campaigns to continuously evolve based on live engagement metrics and accurate pre-aggregated data points relevant to advertisers. We also aim to train the SynFusion hybrid model on diverse, non-synthetic marketing datasets for stronger and more natural generalization across industries and demographic groups. On the frontend, we intend to implement personalized dashboards for both individuals and organizations, with dynamic comparisons between ongoing and past campaigns and visualizing more data in easily digestible formats.

Built With

Share this project:

Updates