pdTetteh/west-africa-voice-support-agent

Voice-first support agent for low-resource West African customer-service settings, with local ASR, hybrid retrieval, grounded answers, and confidence-aware escalation. # West Africa Voice Support Agent A production-minded voice support agent for low-resource customer support scenarios in West Africa. This project demonstrates how to build a safer, grounded support workflow that accepts audio or text, retrieves evidence from a support knowledge base, generates concise responses, and escalates risky or unsupported cases to a human agent. ## Why this project matters Many customer support environments in African markets operate under constraints that are often missing from mainstream AI demos: - multilingual and code-switched user input - noisy audio conditions - informal phrasing and spelling variation - low-bandwidth or unstable connectivity - trust-sensitive financial and account-related issues - limited tolerance for unsupported or risky responses This repo is a focused demonstration of how to design a support assistant for that setting. ## What this project shows This system implements a narrow but realistic support pipeline: 1. Accept voice or text input 2. Transcribe audio locally 3. Normalize the user query 4. Retrieve relevant support documents 5. Generate a response grounded in retrieved evidence 6. Estimate confidence and detect risky cases 7. Either return an evidence-backed answer or recommend escalation The goal is not open-ended chatting. The goal is reliable support behavior in settings where grounding, fallback logic, and escalation matter. ## Current evaluation snapshot - **Top-1 retrieval accuracy:** 83.3% - **Evidence recall@3:** 100.0% - **Average gold coverage:** 84.7% - **Escalation accuracy:** 75.0% - **Average confidence:** 0.771 - **Unsupported answers:** 0 ## Example supported intents - send money failed - cash-out issue - account locked - wrong recipient - KYC / identity verification help ## Core features - **Voice and text support** - **Local offline transcription with `faster-whisper`** - **Hybrid retrieval** - domain-aware lexical retrieval - TF-IDF retrieval - hybrid reranking - **Ground …