DealAgent's live dashboard — 4 companies already researched and stored in the Neo4j knowledge graph
5 autonomous AI agents activate instantly — Tavily, OpenAI, Reka AI, Neo4j, and Airbyte all firing in parallel
All 6 agents completed successfully — from web research to knowledge graph storage in under 60 seconds
Company Intelligence Overview — decision maker identified, top pain point extracted, sales angle ready from live web sources
Reka AI Sales Strategy — personalized playbook with opening hook, value propositions, and objection handling
3-Step Email Sequence — personalized cold outreach, followup, and close emails ready to send in one click
Neo4j Knowledge Graph — all 4 companies connected to their pain points and sources, growing smarter every search
Sidebar knowledge graph — pain points and sources stored persistently, building institutional sales intelligence over time

DealAgent — Autonomous Sales Intelligence

Inspiration

Every sales rep knows the pain. You get a list of 50 companies to reach out to. You spend hours Googling, reading LinkedIn, trying to understand pain points, writing emails, rewriting them, and half the time they don't even reply.

The average B2B sales rep wastes 2,000 hours a year on manual research. That's 40% of their entire work week — gone. Not selling. Just researching.

We asked ourselves: what if 5 AI agents could do all of that in 60 seconds?

What We Built

DealAgent is a fully autonomous sales intelligence system. You type one company name. Five specialized AI agents collaborate in real time to deliver:

Real-time company research from the live web
Decision maker identification and pain point analysis
Personalized sales strategy with objection handling
3-email outreach sequence ready to send
Neo4j knowledge graph that gets smarter every search

How We Built It

We built DealAgent using 5 sponsor platforms working together:

Tavily — Real-time web search API powering our Research Agent
Neo4j Aura — Graph database storing companies, pain points, and relationships as connected nodes
Reka AI — Powers our Strategy Agent for personalized sales playbook generation
Airbyte — Data pipeline keeping our knowledge store synchronized in real time
OpenAI GPT-4o — Drives prospect analysis and personalized email writing
Render — Live cloud deployment

Built with Python and Streamlit for the frontend, with each agent as an independent module that passes results to the next.

What We Learned

Multi-agent orchestration requires careful state management — each agent depends on the previous one's output
Neo4j knowledge graphs compound in value over time — the more companies you research, the more patterns emerge
Forcing structured JSON output from LLMs is critical for reliable agent pipelines
Real-time web data (Tavily) makes the difference between generic and genuinely useful intelligence

Challenges

Agent reliability — Getting each agent to return consistent structured data required careful prompt engineering and JSON output enforcement
State management in Streamlit — Preventing agents from re-running on every UI interaction required session state guards
Neo4j integration — Building meaningful graph relationships that actually reveal patterns took iteration
Time — Building a full multi-agent system with 5 integrations in one day was intense!