Agentic-Reach — Autonomous Multi-Agent Sales Intelligence
Deploys a collaborative swarm of 4 specialized AI agents (Scout, Psychologist, Scribe, Mirror) working in an adversarial loop to generate hyper-personalized sales outreach grounded in real-time Google Search signals.
Overview
Official Repository: Agentic-Reach on GitHub
Built as a submission for the Google PromptWars x Ascent Hackathon, Agentic-Reach is an autonomous multi-agent platform designed to solve the noise problem in digital outreach. Standard sales outreach tools focus on volume, resulting in low reply rates and degraded email sender reputations.
Agentic-Reach tackles this by deploying a collaborative Sales Pod of 4 specialized AI agents working inside an adversarial debate chamber (the "War Room"). They perform live prospect analysis, personality profiling, copywriting, and skeptical critique to produce a single, highly refined outreach message.
Technical Valuation: Replacement Cost: ₹18.5 Lakhs | Hackathon Winner Architecture. Real-time streaming via Server-Sent Events (SSE).
Multi-Agent Architecture: The Sales Pod
The platform splits tasks across 4 distinct agentic personas, collaborating sequentially and cyclically:
Prospect Input
│
▼
┌─────────────┐ Google Search ┌──────────────────┐
│ 🕵️ Scout │ ─── Grounding ────► │ Research Report │
└─────────────┘ └────────┬─────────┘
│
┌────────▼─────────┐
│ 🧠 Psychologist │
│ DISC Profiling │
└────────┬─────────┘
│
┌────────▼─────────┐
│ ✍️ Scribe │
│ Initial Draft │
└────────┬─────────┘
│
┌────────▼─────────┐
│ 🎭 Mirror │
│ Adversarial │◄─── Cynical Prospect
│ Critique Loop │ Simulation
└────────┬─────────┘
│ (Rewrite)
┌────────▼─────────┐
│ ✍️ Scribe │
│ FINAL VERSION │
└──────────────────┘
- 🕵️ Scout (Gemini 2.5 Flash): Conducts real-time company and prospect research using Google Search Grounding to identify active business triggers (funding, hires, launches).
- 🧠 Psychologist (Gemini 2.0 Flash): Analyzes the prospect's public footprint to construct a DISC personality profile (Dominance, Influence, Steadiness, Conscientiousness) and devises the persuasion strategy.
- ✍️ Scribe (Gemini 2.5 Flash): Integrates the research triggers and DISC advice to compose the initial email draft.
- 🎭 Mirror (Gemini 2.0 Flash): Acts as the Adversarial Critic, simulating a cynical, busy prospect who tries to reject the email. It provides feedback and forces the Scribe to iterate until the copy is bulletproof.
Engineering Moat: Real-Time SSE Debate
To provide visibility into the agentic workflow, I engineered a high-concurrency event stream on the FastAPI backend. The frontend subscribes to Server-Sent Events (SSE), rendering a live "War Room" console where users watch the agents debate, critique, and self-correct in real-time.
[System] Initializing session for Sundar Pichai at Google...
[Scout] Deep searching for business triggers via Google Search Grounding...
[Scout] Research complete. 3 business triggers identified.
[Psychologist] Analyzing behavioral patterns and DISC profile...
[Psychologist] Profile: High-D executive. ROI-first, peer tone.
[Scribe] Drafting initial outreach message...
[Mirror] REJECTED: Too generic. No concrete proof points.
[Scribe] Self-correcting... Final version optimized. ✓
Technology Stack
- Large Language Models: Gemini 2.5 Flash (for high-reasoning copy) & Gemini 2.0 Flash (for fast adversarial critique).
- Grounding Engine: Google Search Grounding API (native integration for real-time validity).
- Backend: FastAPI (Python) utilizing async event generators for streaming.
- Frontend: Next.js 15 (React 19) styled with premium brutalist typography.
- Infrastructure: Docker containerization with async request worker threads.