SalesForge Agent

AI sales agent that researches, scores, and writes personalized outreach from raw leads — powered by Elasticsearch hybrid search, ES|QL analytics, and Agent Builder.

Comment

Inspiration

Sales teams waste 40% of their time on manual lead research, qualification, and outreach drafting. They jump between CRMs, LinkedIn, company databases, and email tools — piecing together information that should flow automatically. I wanted to build an agent that does all of this in a single conversation.

What it does

SalesForge is a multi-step AI agent built on Elasticsearch Agent Builder that automates the entire sales intelligence pipeline:

  1. Discover — Analyzes the lead database using ES|QL to surface industry breakdowns and patterns with auto-generated visualizations
  2. Research — Uses hybrid search (BM25 keyword + kNN vector similarity) to find companies matching specific criteria
  3. Score — Applies a transparent, deterministic scoring rubric (0-100) across four dimensions: employee count, funding stage, industry fit, and description quality
  4. Generate — Writes personalized outreach emails referencing specific company details — never generic templates
  5. Explain — Every decision comes with transparent reasoning and a full audit trail

How I built it

  • Seeded 100 synthetic leads with 1536-dim OpenAI embeddings into Elasticsearch
  • Designed hybrid index mappings supporting both BM25 lexical and kNN vector search
  • Built a deterministic 4-dimension scoring rubric (not LLM-based scoring — fully explainable)
  • Created ES|QL query templates for analytics and reporting
  • Configured the agent in Kibana Agent Builder with custom system prompt and 8 tools
  • Used Elastic Workflows for scoring automation and action logging

Challenges I ran into

Built With

Share this project:

Updates