Audio Supporting Agent

About this Gig
Built a modular, end-to-end voice-based AI customer support agent using a full STT → LLM → TTS pipeline. The system accepts spoken customer queries, processes them intelligently using a Retrieval-Augmented Generation (RAG) model, and responds with synthesized speech — replicating a real-world AI support bot.
Requirements
Client Information Checklist 1. 🎯 Business & Use Case What is the product/service your support agent will handle? What language(s) should the agent support? (English only or multilingual?) What are the most common customer queries you receive? 2. 🧠 Knowledge Base Content Can you provide your FAQ documents, product manuals, return policy, pricing info, etc.? How many support topics/documents do you have? Do you want the agent to only answer from documents or also use general AI knowledge? 3. 🎤 Speech-to-Text (STT) Preference Do you have an existing STT API key? (Deepgram, AssemblyAI, Azure, Google) Or should I use a free local model (Whisper)? What is the expected audio input format? (microphone, uploaded WAV/MP3, phone call?) 4. 🔊 Text-to-Speech (TTS) Preference What voice style do you prefer? (Male/Female, formal/friendly tone) Do you have an ElevenLabs or OpenAI TTS API key? Or should I use the free Edge TTS option? 5. 🤖 LLM / AI Model Do you have an OpenAI / Anthropic / Gemini API key? Or should I set it up with a free local model (Ollama)? What tone should the agent use? (Professional, casual, empathetic?) 6. 🌐 Deployment & Integration Where should the agent be deployed? (Web app, mobile app, call center, internal tool?) Do you need it as a REST API that plugs into your existing system? Should it have a chat UI or just the backend pipeline? 7. 💰 Budget & Constraints Are you open to paid APIs or do you prefer fully free/local setup? Any data privacy concerns? (Some clients don't want audio sent to cloud APIs) Expected number of users / call volume per day? 8. ⏱️ Timeline What is your expected delivery date? Do you need testing/demo sessions during development?