AlphaScope - Agentic Equity Research Engine

Generates structured BUY/HOLD/SELL reports for any publicly traded company using an 11-agent LangGraph pipeline running locally on your GPU.

Architecture

User Query > POST /analyze > LangGraph Pipeline > GET /report/{id}

Pipeline (11 Agents):
  planner_agent > (market_data, fundamentals, competitor, industry, news) [parallel]
                      |
                  sentiment_agent <---- news_agent
                      |
              (all converge) > valuation_agent > risk_agent > critic_agent
                      |
              critic_loop (retry failed agent) > report_agent > FINAL REPORT

Prerequisites

• Python 3.11+ (Download)
• NVIDIA GPU with CUDA 12.4 drivers (check with nvidia-smi)
• ~5 GB VRAM recommended (Q4_K_M quantization fits on 6 GB cards like RTX 4050)
• Redis (for caching - install from Redis for Windows)
• ~9 GB free disk space (model + dependencies)

Setup

Step 1: Create virtual environment

cd AlphaScope
python -m venv venv
.\venv\Scripts\activate

Step 2: Install dependencies

pip install --prefer-binary -r requirements.txt

Pre-built CUDA 12.4 wheels are configured in requirements.txt - no source compilation needed.

Step 3: Configure environment

copy .env.example .env

Edit .env only if you need to change defaults or add API keys. The defaults work out of the box.

Step 4: Start Redis

Download and install from Redis releases, then run redis-server.

Step 5: Run the app

python -m app.main

On first launch the app will automatically download the ~4.4 GB GGUF model to app/models/. This only happens once - subsequent launches reuse the cached model.

Step 6: Open the UI

Open http://localhost:8000 in your browser.