ORKA

ORKA Is the distributor of prediction market alpha using AI agents & MCP.
Users get AI analysis on every bet: they choose their LLM (e.g. OpenAI, Claude, DeepSeek) and data sources (e.g. X, Reddit, finance). The system feeds market context into the model so they make better, data-backed decisions and lose less to bad trades.
ORKA also runs a custom MCP marketplace where experienced polymaths deploy trading agents. Those agents’ performance is visible; small-scale traders can subscribe to the ones with strong profits and earn by following their alpha instead of building strategies from scratch.
So ORKA does two things: it puts AI + data on every prediction-market bet so users choose better, and it distributes prediction-market alpha by letting experts monetize agents and letting everyday traders subscribe and share in that edge.

ORKA is built as a multi-service stack with one frontend and several backends. The frontend is a Next.js 16 app (React 19, TypeScript, Tailwind) that serves the UI and exposes Next.js API routes that proxy to backend services and external APIs.
The AI analysis layer is a FastAPI (Python) service (uvicorn on port 8000) with /api/models and /api/chat. It supports multiple LLMs (OpenAI, Anthropic Claude, DeepSeek via config and env keys). The chat endpoint accepts market context, risk tolerance, amount, and optional data sources (e.g. X, Reddit) and injects them into a system prompt before calling the chosen provider.
The Kalshi integration is a separate Express (Node/TypeScript) server on port 4000 that wraps the Kalshi API (markets, events, orderbook, orders, trades). The frontend calls it only through Next.js proxy routes.
Polymarket is integrated via Next.js API routes that call Polymarket’s Gamma API, with an optional Cloudflare Workers proxy for CORS and proxying. A Bun + Hono service in the repo provides Polymarket CLOB/ethers tooling where needed.
The frontend talks only to the Next.js app; all external and backend calls go through these API routes and proxies on the server. That keeps API keys and backend URLs out of the client and gives a single entrypoint for the terminal (markets, AI chat, and future MCP/agent features).

Our project uses the Chainlink CRE (Compute Runtime Environment) SDK to build a trustless order execution workflow that bridges blockchain with external prediction market APIs. The CRE workflow is triggered via HTTP and routes incoming trade requests to either Kalshi or Polymarket. For Polymarket, the workflow validates that the target market is open using Chainlink's HTTPSendRequester, retrieves the wallet private key from CRE Secrets, derives API credentials through L1 authentication with EIP-712 signatures, constructs and signs the order on-chain, and submits it to Polymarket's CLOB — all within a single workflow execution. For Kalshi, the workflow constructs authenticated limit orders and submits them to the Kalshi Trade API. Every external HTTP call in the workflow uses consensusIdenticalAggregation(), ensuring that all nodes in the Chainlink Decentralized Oracle Network (DON) reach agreement on the API responses before proceeding. This means order execution is verifiable, tamper-proof, and not reliant on any single centralized server. The workflow has been successfully simulated using the CRE CLI.