Inspiration
Inspiration Prediction markets are great at summarizing what people think will happen, but they don’t show how much extra research could improve those predictions. At the same time, analysts and students rarely have unlimited time or money to dig into every dataset or article. Insight per Dollar was inspired by the question, “If an AI agent had a small research budget, how should it spend it, and is that effort actually worth it?”
What it does
Insight per Dollar is a Hex data app that turns a limited research budget into measurable improvements on prediction‑market forecasts. Users type a question, pick a market, and set a research budget in “points”; the agent then chooses which public signals (news, stats, macro data, etc.) to “buy,” enriches the market data, and shows how accuracy and confidence change compared to a free‑only baseline. The app visualizes free vs. funded forecasts side by side and explains, in natural language, which research steps were worth the points.
How we built it
We built Insight per Dollar as a Hex notebook that we published as an interactive data app. Python and SQL cells pull prediction‑market data from the Hex‑a‑thon resources, generate candidate public signals, and assign each one a relevance and cost score. A budget allocator module selects the best signals under the user’s budget, feature‑engineering code enriches the market dataset, and simple forecasting models compute free vs. funded predictions, which feed into Hex charts, parameter widgets, and AI narrative cells in the app view.
Challenges we ran into
The first challenge was designing a budget that feels realistic but still uses only publicly available data—no real payments or private APIs—while keeping the math simple enough for a hackathon timeline. Another challenge was keeping the Hex notebook responsive as we chained multiple data sources and models; we had to carefully cache intermediate tables and limit the number of heavy external calls so the app stayed fast and interactive.
#Accomplishments that we're proud of
We’re proud that Insight per Dollar feels like a true agent rather than a static dashboard: it actually makes decisions about which research to do and logs every choice. We also like that the app explains its reasoning in plain language, so non‑technical users can see not just that funded forecasts are better, but why certain news or datasets were chosen over others.
What we learned
We learned how to structure a Hex project so that the same notebook powers both serious modeling work and a polished, shareable app. We also gained a better understanding of how prediction markets behave and how adding structured external information can improve or sometimes fail to improve their signals—teaching us that more data is not always better unless it’s selected thoughtfully.
#What's next for Insight per Dollar
Next, we want to plug in more diverse public datasets (social sentiment, on‑chain indicators, or polling data) and experiment with smarter budget‑allocation strategies, like reinforcement learning or bandit algorithms. Longer term, Insight per Dollar could become a general “research planner” for any analyst in Hex, suggesting the most valuable next dataset or query to run when time and compute are limited.
Log in or sign up for Devpost to join the conversation.