SamuelDasaolu/civic-acess
Domaine:
natural language processingdigital infrastructureType de record:
softwareproject
Créateur:
Sam
Hôte:
*Democratizing access to Nigerian Law using a RAG built on `NCAIR1/N-ATLaS`, a Multi-lingual LLM. Speaks Pidgin, Yoruba, Hausa & Igbo.*
-----
# 🏛️ Civic-Access: Democratizing Legal Knowledge with AI
### *Project by Team SABILAW*
> **🔴 Live Demo:** n-atlas-hackathon.web.app
> **🧠 Legal Engine API:** `civic-backend-528126792252.…`
> **📊 Live Evaluation Logs:** `civic-backend-528126792252.…`
*(Check here to see the "Judge" grading answers in real-time)*
> **🔐 Auth Engine API:** `partner-api-528126792252.us…`
-----
## 📖 Executive Summary
**Civic Access** is a multi-lingual, AI-powered legal assistant designed to bridge the gap between complex Nigerian laws and everyday citizens.
The Nigerian legal system is written in complex English, effectively excluding millions of citizens who speak Pidgin, Yoruba, Hausa, or Igbo. **Civic Access** solves this by combining **Retrieval-Augmented Generation (RAG)** with a **Hybrid Translation Layer**, allowing users to ask questions like *"Wetin be the law for tiff?"* and receive accurate, legally cited answers in their local dialect.
**Core Resources:**
* 📜 **1999 Constitution of the Federal Republic of Nigeria (as amended)**
* 👮 **The Nigeria Police Act 2020**
* 🏠 **2011 Lagos State Tenancy Laws**
-----
## 🏗️ Microservices Architecture
Civic Access utilizes a decoupled **Microservices Architecture** hosted on **Google Cloud Platform (GCP)** to ensure scalability, fault tolerance, and separation of concerns.
### The Ecosystem
1. **Frontend (The Face):** A React/Vite SPA hosted on **Firebase Hosting**. It manages the UI, language selection, and routes API calls to the appropriate backend.
2. **Service A: The Legal Brain (RAG Engine):** A high-performance FastAPI service on **Cloud Run**.
* **Translation:** Google Cloud Translate (v2) performs query-side translation to English. *Disclaimer: This pre-processing is required solely to align user intent with our English-based Vector Store for accurate retrieval, distinct from the N-ATLAS model which performs the fina …