Summarization Service

Generates comprehensive investment analysis reports by extracting financial data from Supabase and feeding it to the Moonshot (Kimi) API.

Architecture

Supabase (documents, earnings, price_snapshot)
    │
    ▼
fetcher.py  ──  Extract & organize data per ticker
    │
    ▼
prompts.py  ──  Build structured prompts
    │
    ▼
summarizer.py  ──  Call Moonshot API (OpenAI-compatible)
    │
    ▼
output/{TICKER}_{DATE}.md  ──  Markdown analysis reports

Prerequisites

• Python 3.10+
• A Supabase project with pgvector enabled and populated tables (documents, earnings, price_snapshot, tracked_tickers). See data-pipeline/pipeline/schema.sql for the DDL.
• A Moonshot API key from platform.moonshot.cn

Installation

cd Summarization
pip install -r requirements.txt

Configuration

Copy the example env file and fill in your credentials:

cp .env.example .env

Required variables:

Variable	Description
SUPABASE_DB_URL	PostgreSQL connection string for Supabase
MOONSHOT_API_KEY	Moonshot (Kimi) API key

Optional variables:

Variable	Default	Description
MOONSHOT_MODEL	kimi-k2.5	Model to use (128k+ context recommended)
MOONSHOT_BASE_URL	https://api.moonshot.cn/v1	API endpoint
MAX_CONTEXT_CHARS	380000	Max chars for prompt context

Usage

# Single ticker
python run.py --ticker AAPL

# Multiple tickers
python run.py --ticker AAPL,MSFT,NVDA

# All active tickers from Supabase
python run.py --all

# Dry run — fetch data only, skip API call (useful for testing)
python run.py --ticker AAPL --dry-run

Reports are saved to output/{TICKER}_{YYYY-MM-DD}.md.

Project Structure

Summarization/
├── run.py              # CLI entry point and orchestration
├── config.py           # Environment-driven configuration
├── fetcher.py          # Data extraction from Supabase + context assembly
├── prompts.py          # System and user prompt templates
├── summarizer.py       # Moonshot API client with retry logic
├── store/
│   ├── __init__.py
│   └── db.py           # PostgreSQL/Supabase data access layer
├── output/             # Generated reports (gitignored)
├── test-script/
│   └── testScriptForVector.py
├── requirements.txt
├── .env.example
└── .gitignore

Data Flow

1. Fetch — fetcher.py queries Supabase for documents (news, 10-K, 10-Q, earnings), structured earnings rows, and price snapshots for a given ticker.
2. Format — Raw data is organized by type and formatted into readable text blocks. If total context exceeds the token budget, older news and filings are truncated.
3. Prompt — prompts.py assembles a system prompt (analyst role + report structure) and a user prompt (all formatted data).
4. Generate — summarizer.py sends the prompt to the Moonshot API with automatic retry on rate limits and connection errors.
5. Save — The generated report is written as a Markdown file to output/.

Database Tables Used

Table	Data Pulled
documents	News articles, SEC filings (10-K/10-Q), earnings text chunks
earnings	Quarterly EPS, revenue, net income, guidance
price_snapshot	Daily close price, P/E ratio, market cap
tracked_tickers	Active ticker list (used with --all flag)