Fharuk/Naija-MarketSense

# Naija-MarketSense 🍅 Naija MarketSense: AI-Powered Price Oracle & Arbitrage Engine Naija MarketSense is a voice-first, multimodal AI agent designed to bridge the information gap in Nigeria's informal food markets. It empowers traders and farmers with real-time price intelligence, arbitrage opportunities, and logistics planning, accessible via natural language (Pidgin English). 🚀 Key Features 1. Voice-First Interface (Accessibility) Problem: Many market traders prefer voice notes over typing. Solution: The app accepts audio input (via st.audio_input), transcribes it using Gemini 2.5 Flash, and extracts intent (Commodity + Market) even from thick Pidgin accents. Audio Response: The AI generates a spoken response (TTS) in a localized accent, advising the user on whether to buy or sell. 2. Arbitrage Engine (Business Intelligence) Problem: Price disparities between markets (e.g., Mile 12 vs. Wuse) are opaque. Solution: The backend MarketOracle scans 6 major Nigerian markets simultaneously. It calculates the Spread (Profit Margin) and visualizes the price differences on a color-coded heatmap, instantly spotting buy/sell opportunities. 3. "Wise Trader" Persona (Context) Problem: Raw data is boring. Solution: The AI persona is prompted to act as a "Wise Market Trader," delivering advice like "Oga, make you carry go Wuse, gain plenty there!" This cultural layer drives user engagement. 🛠️ Technical Architecture Tech Stack Frontend: Streamlit (Python) AI Core: Google Gemini 2.5 Flash (Multimodal capabilities) Voice Synthesis: gTTS (Google Text-to-Speech) Data Visualization: Plotly Express Deployment: Streamlit Community Cloud Agent Workflow Input: User speaks or types: "How much for rice for Bodija?" Perception Agent: Transcribes audio and extracts entities ({"commodity": "Rice", "market": "Bodija"}). Market Oracle: Queries the mock database for the requested item and performs a Full Market Scan for arbitrage. Reasoning Agent: Compares prices, calcula …