Improve Lead Discovery and Email Quality Pipeline

[armaan-71]

Problem

The current pipeline passes raw Google Search results directly to the LLM
for email generation. This produces low-quality leads (blog posts, listicles)
and generic-sounding emails because the LLM only has a 2-line search snippet
as context.

Proposed Solution

Introduce a multi-stage enrichment pipeline:

1. Smart query rewriting — LLM rewrites user queries and selects the
   right SerpApi engine (google vs google_maps) based on query intent
2. Result filtering — LLM classifies results and removes non-company
   pages (blogs, directories) before processing
3. Website scraping — Fetch actual company websites to extract real
   content for the LLM
4. Improved email generation — Use full website context instead of
   search snippets for personalized, higher-quality outreach

Sub-Issues

• LLM-Powered Query Rewriting and Search Engine Selection #27 — Smart query rewriting and engine selection
• LLM-Powered Result Filtering and Classification #29 — Result filtering and classification
• Website Scraping and Content Extraction #30 — Website scraping and content extraction
• Improved LLM Prompts for Analysis and Email Generation #31 — Improved LLM prompts for email generation