austinLorenzMccoy/naija-oracle
Domain:
natural language processingRecord type:
software
Creator:
aus
Host:
Naija Oracle is a sophisticated dual-agent LLM system that reads how real Nigerian users think, speak, and choose — then simulates their voice to generate authentic reviews and delivers hyper-personalised recommendations tuned to Nigerian consumer context.
# 🧠 Naija Oracle: Complete LLM Agent System for Nigerian Cultural Intelligence
**The oracle that speaks Naija**
*[LLM agents that simulate Nigerian consumer voices and deliver hyper-personalised recommendations]*
---
*"No be say the product bad — the vibe no catch me."*
*A Naija Oracle system would predict exactly this sentiment for the right user.*
## 🌟 Overview
**Naija Oracle** is a sophisticated dual-agent LLM system that reads how real Nigerian users think, speak, and choose — then simulates their voice to generate authentic reviews *and* delivers hyper-personalised recommendations tuned to Nigerian consumer context.
The name is deliberate: an *oracle* knows what you'll say before you say it. "Naija" anchors the system in Nigerian cultural specificity — a kind of Pidgin-inflected, context-aware voice that generic recommendation systems erase entirely.
## 🚀 Live Platform
**Frontend**: naija-oracle.netlify.app \
**Backend API**: naija-oracle.onrender.com \
**DagsHub Repository**: dagshub.com \
**Docker Compose**: Fully containerized multi-service setup
## 🎬 Demo Video
**▶ Watch the full demo on YouTube** — live walkthrough of Task A (review simulation with Nigerian Pidgin output and voice playback), Task B (contextual recommendations), cold-start onboarding, and the persona voice fingerprint system.
## 🎯 Problem Solved
Existing LLM-based review and recommendation systems suffer from three compounding failures when applied to Nigerian users:
1. **Cultural voice erasure** — models produce sanitised Standard English that sounds nothing like how Nigerian users actually write reviews
2. **Static user modelling** — users are bucketed into fixed profiles rather than treated as dynamic agents shaped by context
3. **Cold-start ignorance** — Nigerian product categories are severely underrepresented in training data
**Naija Oracle** solves all three with authentic cultural in …