Project Description I am building DrivePK BOLO, an AI-powered urdu voice assistant for a Pakistani car marketplace. The application already exists and is deployed on AWS Elastic Beanstalk, but it now needs professional finishing, stabilization, and future-ready architecture. This is NOT a new app and NOT a rebuild. This project is about improving, stabilizing, and extending an existing codebase in a clean and scalable way. I am looking for a professional freelancer from Pakistan only who can understand long-term product thinking, not quick hacks. Current Working App (For Testing & Understanding) You can test the current version of the app here to understand the existing flow, UI, and limitations: https://bolo.drivepk.com (Test especially on mobile: voice input, Post Car flow, and preview behavior) Core Vision of BOLO (Must Read Carefully) BOLO works as a urdu voice + AI assistant with two intelligence layers: Layer 1 — Predefined / Simple Responses (Audio-Based) For simple and repetitive questions, BOLO should NOT call OpenAI. Instead, it should instantly play pre-recorded MP3 audio. Examples: “How to post a car?” “Seller number is hidden” “Opening DrivePK listing” “Please allow microphone access” Voices / Personalities: Kiran Mahnoor Boss Umar (Up to 10 different voices in future) Voice system must be: Modular Easy to add/remove voices later Fast and reliable Layer 2 — Intelligent / Complex Responses (OpenAI GPT-4) For complex, conversational, or advisory questions, BOLO should: Call OpenAI GPT-4 Generate intelligent responses Optionally convert responses to speech Examples: “Which car is best under 30 lakh in Pakistan?” “Is hybrid better than petrol for city driving?” “Is this price fair for a 2018 Civic?” OpenAI Integration Requirements GPT-4 or GPT-4-class model Backend-only API usage (no frontend keys) Clean prompt design Error handling & fallback to audio responses Cost-safe logic (no unnecessary calls) Architecture must allow future expansion Voice & Speech Requirements Web Speech API for voice input Long dictation support (no auto-stop every few seconds) Auto-restart listening when paused Manual start/stop control Clear listening state indicators Smart routing: Simple intent → MP3 Complex intent → OpenAI Post Car Feature (Very Important) Voice + text posting Raw text → structured parsing Preview before submit Phone OR WhatsApp required (Pakistan 11-digit format) Separate phone & WhatsApp fields Multiple image upload: Gallery Camera Remove individual images Extra notes field (e.g. new tyres, rear camera) MongoDB storage Test database only during development Tech Stack (Must Follow) Frontend: HTML, CSS, Vanilla JavaScript Backend: Node.js, Express AI: OpenAI API (GPT-4) Database: MongoDB Voice Input: Web Speech API Hosting: Temporary full-stack testing (Replit / Render / Railway) Final deployment: AWS Elastic Beanstalk Development & Testing Workflow (Non-Negotiable) You will NOT get AWS access Full app must run on a temporary HTTPS environment I will test everything on mobile Use test MongoDB only Final delivery must include: AWS Elastic Beanstalk–ready ZIP package.json at root No secrets hardcoded No .env in ZIP I will deploy to AWS myself. Project Phases Code stabilization & cleanup Voice system finalization OpenAI GPT-4 integration Advanced AI behavior & scaling Each phase must be testable and approved before moving forward. Who Should Apply (Pakistan Only) Strong JavaScript debugging skills Node.js & Express experience OpenAI API experience Voice-based web app experience Beginners Rebuilders Copy-paste developers Screening Question (Mandatory) How would you design a system that uses MP3 audio for simple responses and OpenAI GPT-4 for complex questions in the same web app? Final Note This is a serious long-term product, not a demo script. Only Pakistani freelancers with real experience should apply