Signal - Bloomberg terminal for prediction markets

Inspiration

Prediction markets like Kalshi are exploding, but the infrastructure for trading on them is still primitive. In traditional finance, high-frequency traders win on latency and information advantage. In prediction markets, traders are still manually refreshing Twitter and reading news articles to make decisions. By the time they click "Buy," the alpha is gone.

We wanted to close this gap. We asked ourselves: What if we could build a "Bloomberg Terminal" for prediction markets that reads the news and executes trades automatically? That question led to Signal—an institutional-grade platform that turns the chaos of the internet into actionable financial signals.

What it does

Signal is an automated trading engine that monitors real-time data sources, analyzes them using AI, and executes trades on prediction markets without human intervention.

Data Ingestion (Scouts): Users configure "Scouts" to monitor specific data streams (e.g., the FDA's Twitter account, RSS feeds for crypto news, or specific subreddits) via the Yutori API.

AI Analysis (The Brain): When a new piece of information arrives, it is instantly sent to an AI agent (powered by Claude 3.5 Sonnet). The agent analyzes the event in the context of a specific market (e.g., "Will the FDA approve Drug X?") and determines a sentiment, confidence score, and trade recommendation (Buy Yes/Buy No).

Execution: If the confidence threshold is met, Signal connects directly to the Kalshi API to execute the trade in milliseconds, beating human traders to the punch.

The Terminal: Users watch this all happen in a "Control Room" dashboard that visualizes the live feed of intelligence, reasoning, and P&L.

How we built it

We built Signal as a modern, event-driven application:

Frontend: Built with Next.js 16 (App Router) and React 19 for a fast, server-side rendered dashboard. We used Tailwind CSS v4 to create a high-contrast, "financial terminal" aesthetic (dark mode, monospaced fonts, dense data displays).

Intelligence Layer: We leveraged Anthropic's Claude 3.5 Sonnet for the reasoning engine. We engineered strict prompts to force the LLM to act as a hedge fund analyst, outputting structured JSON decisions (action, confidence, reasoning) rather than conversational text.

Data & Execution:

Yutori API: Used to abstract away the complexity of scraping Twitter/Reddit and provide clean event streams.

Kalshi API: We implemented a custom robust client to handle cryptographic signing (SHA256 signatures) for secure trade execution.

Backend Logic: Next.js API Routes serve as the orchestration layer, connecting the scouts to the AI agent and the exchange.

Challenges we ran into

Hallucinations vs. Alpha: Getting an LLM to reliably output trading decisions without "hedging" its words was difficult. We had to iterate heavily on the system prompts to make the AI opinionated enough to trade.

Cryptographic Hurdles: Kalshi's API requires signing every request with an RSA key and a timestamp. Debugging the signature verification process in a Node.js environment took significant effort.

Latency: The value of news decays in seconds. Optimizing the pipeline from Tweet Detected → Claude Analysis → Kalshi Order required parallelizing our API calls and stripping out unnecessary logic.

Accomplishments that we're proud of

End-to-End Automation: We successfully built a loop where a simulated tweet can trigger a real order on the Kalshi exchange without a single mouse click.

The UI "Vibe": We managed to nail the professional trading terminal aesthetic. It doesn't look like a toy; it looks like a tool you'd see on a wall street desk.

Next.js 16 Adoption: We lived on the bleeding edge, using the latest Next.js and React 19 features, which made the application incredibly snappy but required us to solve new types of hydration and server-component errors.

What we learned

"News-as-Code": We learned that unstructured text (news) can be treated as structured code if you have a deterministic AI layer in the middle.

Agentic Workflows: We discovered that single-shot prompting isn't enough for finance. The best results came when we gave the model "context" (current market prices) alongside the news.

Market Microstructure: working with the Kalshi API taught us the nuances of order types (Limit vs. Market) and how liquidity impacts automated strategies.

What's next for Signal - Bloomberg terminal for prediction markets

Backtesting Engine: allowing users to run their strategies against historical news data to verify win rates before deploying real capital.

Multi-Agent Debate: Instead of one AI deciding, having a "Bull" agent and a "Bear" agent debate the news, with a "Manager" agent making the final call.

More Markets: expanding beyond Kalshi to Polymarket and other event contracts.

Websockets: Upgrading from polling to full websocket connections for sub-second latency.