Prediction Assistant

project-prediction-assistant Project Page

active project-page

Vision

AI-powered trading assistant for Kalshi prediction markets. Uses automated strategies (watchdog scanning, portfolio stacking with Option D hedging) to generate consistent returns while bounding worst-case loss. The system deploys a 4-bot architecture — Pregame Stacker, Late-Game Lock, Bulk Sweep, and Edge Learner — driven by a watchdog fan-out pattern that scans markets and dispatches opportunities to specialized bots. Users connect their Kalshi API credentials through an iOS app at prediction-assistant.com and the system trades on their behalf — they watch results live in the official Kalshi app.

User Stories

Key Story Note Role Success Metric
portfolio-builder <a href="story-prediction-assistant-portfolio-builder">AI Portfolio Builder</a> Trader (Lucas) Positive EV across 100+ trades, worst-case loss under 6%
watchdog-trading <a href="story-prediction-assistant-watchdog-trading">Watchdog Trading</a> Trader (Lucas) 85%+ win rate, $0.10–$0.15 avg profit per contract
app-experience <a href="story-prediction-assistant-app-experience">App Experience</a> Consumer First automated trade within 5 min of connecting credentials
credential-onboarding <a href="story-prediction-assistant-credential-onboarding">Credential Onboarding</a> Consumer 90% of users complete setup without support
platform-setup <a href="story-prediction-assistant-platform-setup">Platform Setup</a> Developer (Lucas) CI/CD push-to-deploy in under 10 minutes
landing-page <a href="story-prediction-assistant-landing-page">Landing Page &amp; Registration</a> Consumer 80% of visitors who start registration complete it
bot-marketplace <a href="story-prediction-assistant-bot-marketplace">Bot Marketplace</a> Consumer User activates first bot within 2 minutes of login

Architecture

  • Domain Model — User, Credential, Strategy, Portfolio, Trade, Market entities
  • Data Flow — watchdog scan loop, position monitoring, user auth flows
  • Deployment — k3s + Tailscale Funnel + Hetzner edge + Caddy for prediction-assistant.com
  • Rails — Rails 8 application architecture
  • API — API layer design
  • App — Application structure
  • Postgres — Database schema and design
  • Frontend — Frontend architecture
  • Auth — Authentication and authorization
  • Keycloak — Keycloak integration for prediction-assistant
Key decisions:
  • Rails 8 with Solid Queue (no Sidekiq/Redis)
  • Kalshi API auth: RSA-PSS SHA-256 per-request signatures
  • Strategy polymorphism via strategy_type + jsonb parameters
  • Three-repo model: pal-e-platform, pal-e-services, pal-e-deployments

Board

Status

Project setup complete. All 16 tickets reviewed and in todo across 4 sprints. Bot architecture established (4 bots on watchdog fan-out). Strategy docs, API integration, and bot specs documented. PR #19 merged with bot specs and user stories. PR #20 merged with Late-Game Lock math fix and Edge Learner synthetic parlay clarification. Rails scaffold and infrastructure provisioning are next (Sprint 1).

Milestones

  • 2026-07-02 — Strategy research and API exploration complete
  • 2026-07-03 — Bot architecture established, all tickets reviewed and moved to todo, PR #19 and #20 merged

Repos

Repo Platform Role Status
ldraney/kalshi-assistant Forgejo Rails app + docs Active (docs only, Rails scaffold pending)
ldraney/kalshi-assistant-ios Forgejo Turbo Native iOS shell Not yet created