Browser Automation & AI Reply System
Publicada el 2026-07-27
Descripción de la oferta
I need a production-ready system that can run scheduled tasks across many AdsPower browser profiles, each profile locked to its own residential or ISP proxy so sessions stay consistent. The automation itself can be built with Playwright or Puppeteer, whichever you are faster with, but it must launch through the AdsPower API, pick the correct proxy, perform the scripted actions, then hand control back to a Celery/Redis job scheduler for the next run. On top of that engine, I want an AI microservice that listens for new customer messages, enriches the prompt with live inventory and pricing pulled from Postgres, and sends an appropriate reply through an LLM (OpenAI or Anthropic). Accuracy matters more than creativity here: the reply must reference the customer’s question and the current data in our database. Key points you’ll tackle • AdsPower integration for profile creation, proxy assignment and session handoff • Reliable proxy management (sticky sessions, geo-targeting, rotation logic) • Automation scripts in Playwright/Puppeteer with robust error handling and logging • Celery/Redis scheduling so each profile’s jobs run on a defined cadence without overlap • LLM API calls that generate responses to customer messages and queries, using real-time inventory and pricing pulled via SQL • Postgres schema and queries to expose that business data quickly • Optional but welcome: webhook flow with the GoHighLevel API so messages and replies sync automatically Deliverables 1. Source code for the browser automation module 2. AI reply microservice with environment-configurable keys and endpoints 3. Docker-compose or similar so the whole stack spins up reproducibly 4. README that shows how to add new automated actions, set proxy rules and extend the AI prompts 5. A short recorded demo proving multiple AdsPower profiles run concurrently and the AI returns contextual replies Acceptance criteria • At least 15 concurrent AdsPower sessions run without collision or shared cookies • Average reply latency from message receipt to AI-generated response under 15 sec • Logged success/error metrics visible in Grafana or a simple dashboard • Clean hand-off so future devs can swap AdsPower for GoLogin with minimal code changes If you have already built something similar and can show code or a live example, that will put you at the top of the list.
Skills
Fuente original: freelancer