PreTerm

## Inspiration

We built PreTerm to make prediction markets easier to understand and act on. Most tools show price movement, but they do not help users connect contracts to headlines, macro signals, planning decisions, and research context in one place. We wanted a workstation that turns scattered information into a clearer picture.

## What it does

PreTerm is a prediction-market workstation for tracking live binary contracts, reading structured event briefs, mapping headlines to relevant markets, analyzing sentiment, researching stocks and SEC filings, monitoring macro data, saving watchlists and desk views, planning around real-world dates, configuring alerts, and chatting with a context-aware AI copilot.

## How we built it

We built PreTerm as a full-stack web app with a FastAPI backend and a React + TypeScript frontend. The backend handles authentication, market ingestion, alerting, research APIs, sentiment analysis, planner logic, and AI-powered copilot features. The frontend provides the workstation experience across Monitor, Headlines, Research, Planner, Watchlists, and Settings. We used SQLite for local persistence, designed the app to support live Kalshi-style market data with seeded fallback data for demos, and integrated Gemini, FRED, Yahoo Finance, SEC EDGAR, Reddit/RSS feeds, and optional ElevenLabs voice features.

## Challenges we ran into

One major challenge was normalizing data from very different sources such as live market feeds, macroeconomic series, financial research endpoints, and news/headline inputs into one coherent interface. We also had to handle API rate limits, offline/demo fallbacks, and background refresh performance so the app stayed responsive. Another challenge was making the AI copilot useful and grounded in real workspace context instead of producing generic answers.

## Accomplishments that we're proud of

We are proud that PreTerm feels like a complete workstation instead of a single-feature prototype. It combines live monitoring, research, sentiment, planning, saved workflows, and AI assistance in one product. We are also proud of building resilient fallbacks for demos, grounding the copilot in market context, and shipping a product that is both technically deep and easy to navigate.

## What we learned

We learned that prediction-market users need context as much as they need prices. We also learned how important strong data modeling and fallback strategies are when a product depends on multiple external APIs. On the product side, we learned that AI features work best when they are embedded in a workflow and tied to real user state, not added as a standalone chatbot.

## What's next for PreTerm

Next, we want to deepen the real-time experience with more live market coverage, sharper alerting, and richer historical analytics. We also want to improve copilot grounding, expand portfolio and scenario analysis, strengthen deployment for production use, and make collaboration features possible for teams tracking the same markets together.

Built With

• elevenlabs
• fastapi
• fred-api
• google-gemini
• hmmlearn
• httpx
• kalshi-api
• nltk-vader
• postgresql
• python
• react
• react-router
• recharts
• scikit-learn
• sec-edgar
• sqlalchemy
• sqlite
• typescript
• vercel
• vite
• yfinance