I want to layer in an AI backbone to my classified ads website that makes every search sharper and every ad better. Concretely, the project breaks down into two core tracks. 1. Search intelligence The engine must read natural-language queries (“red sofa under $300 near me”) and return truly personalized results based on a shopper’s history and behaviour. Vector-based search, embeddings, or fine-tuned LLMs are all acceptable so long as latency stays low and relevance measurable. 2. Ad composition When a seller starts a listing, the AI should auto-draft the title, body copy and tags with a clear bias toward search-engine optimisation. I expect keyword-rich phrasing, correct categories, and suggestions that nudge the seller to complete missing attributes. Because we’re taking the AI plunge, I’d also like you to wire up three ancillary capabilities: • User recommendations that surface “you may also like” listings in real time • Fraud detection scores to flag suspicious posts before they go live • Automated, conversational customer-service replies for common queries Deliverables will be accepted when: • Search results can be A/B tested against our current baseline and show uplift in CTR or conversion. • Generated ad copy passes a basic SEO audit tool and can be published without manual rewrites. • Recommendation, fraud and support modules expose clean REST or GraphQL endpoints, documented and covered by unit tests. • Deployment scripts (Docker, Kubernetes or similar) reproduce your environment and models in staging. I’m agnostic about frameworks—Python, Node.js, TensorFlow, PyTorch, OpenAI, or Elasticsearch are all on the table—provided the code is maintainable and model choices are justified.