CultureHedge

Webpage
QuestDB
User Page
Metamask Transaction
Temporal
Blockchain details

Inspiration

CultureHedge was inspired by the idea that markets are influenced not only by fundamentals, but also by culture, sentiment, public attention, and narrative. Topics trending across news, social media, and public discussion often shape market behavior before traditional financial signals fully catch up.

For the Finance Track, we wanted to build a product that connects current cultural events with financial analysis in a meaningful way. That led us to the idea of an attention market where users can track topics, trade on their momentum, and understand the narrative behind market movement.

What it does

CultureHedge is an AI-powered attention market platform.

Admin users can add new topics into the system.
Lower-level research agents gather fresh information about those topics from sources like GDELT and Tavily.
The incoming data is stored over time and used to generate both market movement and narrative summaries.

For normal users, the platform allows them to:

authenticate and enter the marketplace
view active topics
buy or sell positions
view wallet information
read narrative explanations for market movement
get portfolio-level insights

The system is powered by three main agents:

Market Agent — drives market behavior for each topic
Narrative Agent — summarizes recent happenings and trends
Portfolio Agent — analyzes the user’s holdings and explains possible risks or opportunities

How we built it

We built the frontend using React and TypeScript, with authentication and wallet-connected flows for users entering the marketplace. The backend is built with FastAPI, which acts as the main API layer between the frontend and the agent ecosystem.

At the core of the system is a multi-agent orchestration layer inspired by AGNTCY-style coordination. We used Temporal to schedule and control recurring workflows so that lower research agents can fetch fresh information regularly.

These research agents collect data from external APIs such as:

GDELT
Tavily

We stored time-series observations in QuestDB so we could track how attention, sentiment, and narrative signals evolve over time. We also used PostgreSQL for structured application data. For the market and wallet side, we integrated a lightweight blockchain setup instead of using something heavier like Hyperledger, allowing users to hold and manage positions in a decentralized way.

Challenges we ran into

One of the biggest challenges was dealing with noisy real-world data. Cultural and public signals are messy, fast-changing, and difficult to structure. Converting that data into something useful for market behavior was not straightforward.

Another challenge was coordinating multiple moving parts at once. We were working with:

agents
APIs
recurring workflows
multiple databases
a frontend
a backend
a blockchain layer

Making sure all of these pieces communicated properly and stayed consistent required a lot of careful design.

We also found it challenging to make the system explainable. We did not want users to only see a topic moving up or down. We wanted them to understand the story behind that movement, while still keeping the product intuitive and usable.

What we learned

We learned that specialized agents work much better than trying to make one agent do everything. Splitting responsibilities across research, market reasoning, narrative generation, and portfolio explanation made the system more modular and easier to reason about.

We also learned that time-series thinking is essential for this kind of product. With cultural and sentiment-based signals, the value is not just in the latest data point, but in how that signal changes over time.

Most importantly, we learned that users need explanation alongside action. A market interface becomes much more useful when it is paired with narrative context and portfolio-level reasoning.