Skip to content

Latest commit

 

History

6 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 

Repository files navigation

AlphaBoundary

Build, validate and deploy quantitative trading strategies — without writing code.

Professional-grade quantitative research platform powered by local-first computing, statistical validation and native execution.

🌐 Website · Documentation · Roadmap · Join Beta


Private Beta Windows Docker Binance Local Compute AI Copilot


The Research Workflow

At the heart of AlphaBoundary is a single, continuous research loop. Every stage feeds the next. Nothing is optional. Nothing is skipped.

        Idea
         |
         v
 Feature Engineering
         |
         v
     CALIBRATE
     (Bayesian TPE · Walk-forward · IC screening · Regime detection)
         |
         v
 Monte Carlo Simulation
         |
         v
    Stress Testing
         |
         v
   Validation Report
         |
         v
       Backtest
   (Real-cost fills · Slippage · Funding)
         |
         v
 Deploy (Paper → Demo → Live)
         |
         v
     Execution
         |
         v
  Drift Detection
         |
         v
    RECALIBRATE
    (When drift exceeds threshold)

Calibration is the core. Every strategy is Bayesian-calibrated across multiple walk-forward windows, with full IC screening and regime-adaptive parameter tuning, before it can be deployed. The backtest is a consequence of good calibration — not a substitute for it.


Screenshots

Coming soon.


Our Philosophy

Most quantitative platforms focus on one stage of the workflow.

Some stop at backtesting.

Others stop at execution.

AlphaBoundary was designed around a different idea:

A trading strategy should pass a rigorous validation pipeline before reaching live markets.

Research. Validate. Deploy. Monitor. Improve. Repeat.


What is AlphaBoundary?

AlphaBoundary is a local-first quantitative research platform for systematic traders.

It covers the entire strategy lifecycle — from feature engineering to live execution — through a structured, statistically rigorous workflow.

Unlike traditional platforms that focus on either research or execution, AlphaBoundary connects every stage through a single, continuous workflow.

The platform enforces one rule: what you test is what you trade.

Every backtest uses realistic fill models. Every feature is screened for predictive power. Every strategy is validated through walk-forward analysis before deployment.

No shortcuts. No optimistic assumptions. No black boxes.


Why AlphaBoundary?

Capability No-Code Bot Builders Cloud Research Platforms AlphaBoundary
Code required No Yes No
Statistical validation None Partial Full pipeline
Real-cost backtesting No Approximate Exchange-realistic
Live execution bridge Manual Self-built One click
Drift detection No No Built-in
Local compute N/A Cloud-only Native C# agent
AI research assistant No No Quant Copilot

Core Modules

  • Research — Feature engineering, information coefficient screening, correlation analysis, regime detection
  • Calibration — Bayesian TPE optimization, walk-forward validation, parameter stability scoring, regime-adaptive weighting
  • Validation — Monte Carlo simulation, stress testing, deflated Sharpe ratio, robustness analysis
  • Execution — Native execution engine, position management, portfolio allocation, realistic order simulation with slippage and partial fills
  • Monitoring — Strategy health tracking, drift detection, exchange reconciliation, risk engine with kill-switches and daily loss limits
  • Quant Copilot — AI assistant that understands your research context, explains statistical metrics, interprets validation results and guides your workflow

Quant Copilot

Most quantitative platforms give you tools. AlphaBoundary gives you a research partner.

What it does:

  • Explains statistical results in plain English
  • Guides you through the research workflow step by step
  • Interprets validation outputs and flags concerns
  • Assists during live execution with real-time context
  • Answers questions about methodology and platform features

Example interactions:

"Why did this feature fail IC screening?"

"Is this walk-forward result statistically significant?"

"Should I be concerned about this drift alert?"

Privacy model: The Copilot operates on metadata and computed metrics — never on raw market data. Your ticks and positions stay local.


Architecture

    Your Browser
         |
         v
    Cloud API (orchestration + state)
         |
         v
    Desktop Agent (C# native)
         |
         v
    Research Engine (local compute)
         |
         v
      Exchange
         |
         v
   Execution Reports
         |
         v
     Dashboard

Key principles:

  • Heavy compute runs locally — Calibration, IC analysis and regime detection execute on your hardware via a native C# agent. No server bottlenecks. No RAM ceilings.
  • Cloud orchestrates, never computes — The cloud layer manages sessions, stores configurations and synchronizes state. Raw market data never touches our servers.
  • Your data stays yours — Ticks, fills and positions remain on your machine unless you explicitly choose to export them.
  • Event-driven pipeline — Feature updates propagate in real time. No polling. No batching delays.

Desktop Agent

Heavy computations are executed locally through a native C# desktop agent.

  • Runs locally — Your hardware, your rules
  • Native C# — Zero-GC overhead, direct memory structs
  • Docker-ready — Deploy on laptop, workstation or cloud VM
  • Cross-machine deployment — Same agent, same performance, anywhere
  • Multi-core execution — Parallel calibrations with shared state

The agent handles the heavy lifting so the cloud only orchestrates. Result: lower costs, better privacy, no compute limits.


Design Principles

AlphaBoundary is built around a few simple principles:

  • Local-first computing — Heavy math runs on your machine, not our servers
  • Research before execution — No strategy reaches live markets without passing validation
  • Statistical rigor over optimistic backtests — Realistic fills, deflated metrics, walk-forward analysis
  • Realistic execution models — What you test is what you trade
  • Explainability before automation — Every decision is inspectable and attributable
  • AI as an assistant, never a decision maker — Quant Copilot guides, you decide
  • User-owned compute — Your hardware, your performance ceiling
  • User-owned data — Raw ticks and positions never leave your environment

Tech Stack

Layer Technology
Frontend React
Desktop Agent C# (.NET, native structs, zero-GC)
Backend Node.js
Communication gRPC (binary, high-throughput)
Storage Parquet (columnar, compressed)
Database PostgreSQL
Streaming WebSockets
AI Local context-aware LLM (metadata only, no raw data)

Roadmap

Recently Completed

  • Web platform (React + local-first architecture)
  • Desktop agent — Windows (C# native)
  • Binance integration (testnet + live)
  • Core research pipeline (feature engineering → validation → execution)
  • Quant Copilot (context-aware AI assistant)

Current — Private Beta

  • UI polish and responsive refinements
  • Documentation suite
  • Onboarding workflow

Next — Q3 2025

  • Bybit and OKX integrations
  • macOS desktop agent
  • Public beta release

Future — 2026

  • Multi-broker support (10+ exchanges)
  • Portfolio-level analytics
  • Strategy marketplace
  • Team collaboration features

Documentation

  • Getting Started — From zero to first backtest
  • Research Guides — Feature engineering, IC testing, regime analysis
  • Calibration — Bayesian TPE, walk-forward validation, parameter stability
  • Validation — Monte Carlo, stress testing, deflated Sharpe
  • Execution — Paper, demo and live trading
  • Desktop Agent — Installation and configuration
  • Quant Copilot — Capabilities and privacy model
  • API Reference — Programmatic access to all operations

Community

Channel Status
Website Live
GitHub Active
Documentation Planned
YouTube Planned
Discord Planned

The Local-First Quant Research Platform.

AlphaBoundary — Research first. Execution second. Marketing last.

About

Local-first quantitative research platform for building, validating, deploying and monitoring systematic trading strategies.

Topics

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors