Skip to main content

System Architecture

The Legislative Analysis tool is designed as a modern, decoupled web application that leverages GenAI for backend data processing.

High-Level Overview​

The system consists of three main layers:

  1. Data Ingestion & Processing Layer: Python scripts using Google Gemini to analyze PDFs.
  2. Service Layer: A FastAPI backend that serves the structured data.
  3. Presentation Layer: A Next.js frontend that provides an interactive UI.

Component Details​

1. Offline Processing (The "Brain")​

  • Input: Raw PDF files of Sri Lankan Acts.
  • Engine: scripts/update_docs_data.py and ldf/ modules.
  • AI Model: Google Gemini 2.0 Flash (via google.genai).
  • Output: Structured JSON data and a populated SQLite database.
  • Role: This layer runs asynchronously to the web app. It pre-computes the analysis so the web app is fast and responsive.

2. Backend API​

  • Framework: FastAPI.
  • Responsibility:
    • Exposes endpoints for listing acts, searching, and retrieving detailed metadata.
    • Handlers complex queries that might be too heavy for the client-side.
    • Manages database connections via SQLModel.

3. Frontend UI​

  • Framework: Next.js (React).
  • Styling: Tailwind CSS + Shadcn UI.
  • Responsibility:
    • Renders the interactive dashboard.
    • Displays Acts in a responsive Card Grid.
    • Visualizes relationships between acts (amendments, references).

Data Flow​

  1. Ingestion: A new PDF is placed in the data/ folder.
  2. Analysis: The extraction script reads the PDF, sends text to Gemini, and receives a structured JSON response.
  3. Storage: The JSON is parsed and stored in research.db tables (Act, Amendment, Entity).
  4. Serving: When a user visits the web app, the Backend fetches this structured data and sends it to the Frontend for display.