Title: AI-powered lead enrichment — match LinkedIn profiles + websites to accountant database

Description: I have a large CSV database of US accountants (name, company name / DBA, city, state, phone). I need an enrichment pipeline that, for each record: Runs a Google search (Google Custom Search API or a SERP provider — your recommendation) Passes the top 10 results + the record's details (name, company, city) to an AI model that identifies which LinkedIn profile (linkedin.com/in/...) and which website actually belong to that person — and returns a confidence level with a one-line reason Writes the enriched row incrementally to the output The AI matching step is the core of this job. Rule-based parsing fails on common American surnames; I want a cheap, fast LLM (GPT-4o-mini, Gemini Flash, Claude Haiku or similar) doing the disambiguation, with low-confidence rows flagged for review, never guessed. Requirements: I pay for the search API and AI API directly with my own keys — quote me your realistic estimate of both costs per 1,000 records in your proposal Must handle a very large dataset reliably: incremental saves, resume after interruption, rate limiting, retries, failed rows logged separately. Python pipeline preferred; if you propose an automation platform (n8n etc.), explain how it survives hundreds of thousands of iterations without babysitting No direct Google scraping, no logging into LinkedIn Process: validation run on a 1,000-row sample first — I hand-check the matches and we tune the search query + AI prompt together — then the full run. Deliverables: the pipeline + short README (run, resume, where keys go), the enriched CSV (original columns + linkedin_url, website_url, confidence, ai_reason), and summary stats (match rates, low-confidence count).