ZhuLinsen/alphasift

AI-native stock screening engine with full-market discovery, LLM ranking, risk-aware scoring, and auditable evaluation. AI选股

What it solves

AlphaSift is a stock discovery and ranking engine designed to help users and AI agents scan the broad market to find high-potential stocks based on specific strategies. It bridges the gap between raw market data and actionable shortlists by combining deterministic technical filters with optional LLM-based qualitative reasoning.

How it works

The engine operates through a multi-layered pipeline:

  1. L1 Deterministic Screening: Applies hard filters and factor scoring across the full market snapshot to narrow down the universe.
  2. L2 LLM Ranking: (Optional) Uses LLMs to perform structured reasoning on the remaining candidates, analyzing catalysts, risks, and confidence levels.
  3. L3 Post-Analysis: (Optional) Runs final checks via a local scorecard or external analyzers like the Daily Stock Analysis (DSA) service.
  4. Enrichment & Discovery: The system can enrich candidates with daily K-line features (e.g., MACD, RSI) and discover market "hotspots" (trending topics/sectors) with timeline evidence.
  5. Evaluation Loop: Users can save runs and evaluate them later (T+N) to track follow-through, failed breakouts, and drawdowns.

Who it’s for

  • Quantitative Researchers: Those who want to implement and audit auditable YAML-based screening strategies.
  • AI Agent Developers: Developers building financial agents who need a structured, callable interface (SKILL.md) for market discovery.
  • Retail Traders: Users looking for a systematic way to discover value, trend, or momentum stocks without manual scanning.

Highlights

  • Agent-Native: Includes a dedicated skill description and a local JSON API for seamless integration with AI agents.
  • Hybrid Ranking: Combines hard mathematical factors with LLM judgment for a more nuanced selection process.
  • Robust Data Sourcing: Implements a priority-based fallback system across multiple A-share data providers (Sina, Tushare, AkShare, etc.) to ensure uptime.
  • Auditable Strategies: Strategies are defined in YAML, making them easy to modify, share, and audit.
  • Built-in Evaluation: Includes tools to save runs and perform T+N evaluation to measure actual performance against subsequent price action.

Related

  • Project
    witt3rd/oh-my-hermes

    A collection of multi-agent orchestration skills for Hermes Agent that enables deep research, consensus-based planning, and verified execution of complex tasks.

  • Project
    wh000wh000/awesome-jev-live

    Awesome Jev Live is an automatically refreshed directory of public projects that use TypeSafe AI’s Jev System One decision model, providing categorized listings, evidence grades, and a searchable multilingual site.

  • Project
    agenticloops-ai/agentic-ai-engineering

    A hands-on tutorial repo that teaches engineers to build AI agents from scratch using plain LLM APIs, tools, and a loop — no frameworks until you understand the primitives. It progresses from a first API call to production-grade agents, covering patterns, testing, loop engineering, and framework comparisons.

  • Project
    ahmadvh/octochains

    A lightweight Python framework for parallel, isolated multi-agent reasoning and consensus, preventing agent bias and groupthink in high-stakes decision-making.