I’m looking to build a lightweight agentic AI that automatically scouts the web for each prospect on my lead list and compiles clear, up-to-date insights into what that company is likely to care about right now. The emphasis is on customer preferences, not generic firmographics, so the agent should draw signals from publicly available sources such as the company’s own site, recent news coverage, press releases, blog posts, and any other open data it can reach. Core workflow 1. Input: a simple list of company names or domains. 2. Autonomous web navigation and scraping of the sites and articles it finds. 3. NLP-driven extraction of preference indicators—e.g., product lines mentioned, initiatives highlighted, pain points, technology keywords, sustainability statements. 4. Summarisation and scoring so I can see, at a glance, which topics will resonate in outreach. 5. Output: structured JSON/CSV plus a short natural-language brief for each lead. Technical expectations • Python with libraries such as LangChain, BeautifulSoup/Scrapy, GPT-4 (or comparable LLM), and a lightweight vector database (Pinecone, FAISS, etc.) for context retrieval. • Modular code so new sources can be added easily. • Respect robots.txt and include a basic rate-limit. • README with set-up steps, environment variables, and example commands. What success looks like - I feed in 50 domains and receive a file that highlights the top 5 talking points for each company plus the evidence sentences and source URLs. - Average run time per domain stays under two minutes on a standard cloud VM. - The insights are specific enough that a salesperson could craft a personalised email without further research. If you’ve built autonomous agents, lead-gen scrapers, or NLP summarisation pipelines before, your experience will be directly relevant. Let’s create a focused prototype that I can test quickly, then iterate together on enrichment and UI once the core engine proves itself.