BUILD STORY ยท APRIL 2026

How HuntR was built

From idea to deployed product in under 6 hours.

Build timeline

  1. Hour 0

    The Problem

    Most Indian startups can't afford a sales team. Manual B2B prospecting takes 3-4 hours daily and produces generic outreach that gets ignored. We decided to fix that.

  2. Hour 1

    Core Agent Pipeline

    Built 5 specialized Google ADK agents: Scout, Researcher, Scorer, Outreach, and Followup. Connected Serper and Tavily for real-time web intelligence. First end-to-end lead generated in terminal. Configured dual-model setup: Gemini 2.5 Pro for reasoning-heavy agents, Gemini 2.5 Flash for speed-optimized agents.

  3. Hour 2

    Self-Correction + API

    Added Manager Agent orchestration with automatic retry logic. Built FastAPI backend with SSE streaming for real-time agent trace. Connected Brevo SMTP for email delivery.

  4. Hour 3

    Dashboard + Deploy

    Built Next.js dashboard with live agent pipeline visualization. Deployed backend to Cloud Run, frontend to Vercel. First production hunt completed.

  5. Hour 3 - 6

    Intelligence + Polish

    Upgraded Scout with India-specific query strategies. Added Firestore persistence, CSV export, email open tracking, voice input, campaign comparison, and full light theme.

  6. Final

    Production-Ready Product

    Landing page, build story, full responsiveness pass. HuntR is now a production-ready B2B sales automation platform.

Key technical decisions

Why Google ADK over CrewAI

Google ADK integrates natively with Vertex AI Gemini 2.5 and fits this architecture immediately. We use Gemini 2.5 Pro for the Manager and Outreach agents where reasoning quality matters most, and Gemini 2.5 Flash for Scout, Researcher, Scorer and Followup where speed is the priority.

Why SSE over WebSockets

Server-Sent Events are simpler, more reliable, and perfectly suited for one-directional agent trace streaming. No connection management overhead.

Why Firestore over PostgreSQL

Firestore's schemaless nature matched our evolving lead data structure. Zero configuration, auto-scaling, native GCP integration.

Why two Gemini models instead of one

Gemini 2.5 Pro produces noticeably better cold email personalization and orchestration reasoning. Gemini 2.5 Flash is 10x cheaper and fast enough for structured data extraction. Using both optimizes quality where it matters and speed everywhere else.

57+ leads found in testing
163 seconds average pipeline time
5 agents, 0 human intervention
Built in < 6 hours