Home Services About Contact
Architecture

4-agent pipeline.
Zero manual steps.

🔍
Prospector

Finds real businesses from OSM, Google Places, Apollo, Hunter. MD5 deduplication across all sources.

🎯
Scorer

GNN spatial scoring + AI enrichment. Scores leads 1–10 with fit score, pain points, icebreaker.

✍️
Writer

Generates personalized email + WhatsApp per lead. LLM-as-Judge evaluates on 5 dimensions before send.

📤
Delivery

Sends via SendGrid + Gupshup. Auto-schedules Day 3 and Day 7 follow-ups via Celery.

Infrastructure underneath: Each agent communicates via Redis queues with per-job isolation. If the Writer fails mid-job, the Prospector's work is not lost — state is persisted in PostgreSQL. This is the Orchestration Layer in production.
Results

Real data.
Real delivery.

4
Lead sources connected
OSM · Google · Apollo · Hunter
202
SendGrid response code
Real emails delivered to Mumbai businesses
11
AI techniques
Research papers implemented in production
🧠 AI Techniques Implemented
  • OPRO — self-optimizing prompts. Tracks reply rates per prompt version, uses LLM to evolve better prompts weekly. Diversity score prevents prompt collapse.
  • Reflexion — failure learning. Agent tracks what went wrong and adjusts behaviour automatically.
  • MemGPT — tiered memory. Working memory, episodic, long-term. Full conversation memory per lead.
  • ReAct — reasoning traces. Agent explains its decisions before acting.
  • LLM-as-Judge — 5-dimension evaluation (personalization, cultural fit, CTA strength, tone, clarity) before every send.
  • RAG — dynamic few-shots from historical high-performing messages.
  • A/B Testing — framework for comparing prompt and message variants.
⚡ Infrastructure Problems Solved
  • Multi-source deduplication — MD5 fingerprint hashing name + company + email + phone prevents duplicates across 4 sources and across agent runs.
  • Async multi-agent pipeline — 4 agents communicate via Redis queues with per-job isolation. Partial failure recovery — Prospector's work survives Writer failure.
  • GNN spatial scoring — businesses modeled as nodes in a spatial graph. Geographic demand patterns and business clustering produce dramatically better lead quality than flat-list scoring.
  • LLM self-optimization without collapse — OPRO implementation maintains diversity score across active prompt versions to prevent all prompts converging to one style.
🛠️ Full Tech Stack
Python · FastAPI PostgreSQL · SQLite Redis · Celery Claude Sonnet (primary) GPT-4o-mini (fallback) OpenRouter SendGrid Gupshup WhatsApp API Google Places API OpenStreetMap Apollo.io Hunter.io Razorpay JWT Auth GNN Spatial Scoring React SPA Render · Netlify
What we can do for you

Real examples.
Real automations.

AI Agents
Vendor Invoice Processing
500+ invoices processed monthly — discrepancies flagged, Tally updated, zero manual entry.
Review Intelligence
Monitors Amazon & Flipkart reviews daily, alerts you instantly when ratings dip below 4 stars.
Govt Tender Monitor
Scans GeM & CPPP portals daily, filters relevant tenders, delivers a digest straight to your phone.
MIS Report Automation
Raw sales data in, polished management report out — every Monday morning, automatically.
Wholesale Order Workflow
WhatsApp bot handles your entire order-to-dispatch pipeline for distributors end to end.
Competitor Price Tracker
Tracks competitor pricing on Amazon & Flipkart daily, sends you a comparison report every evening.
KYC & Client Onboarding
Collects documents via WhatsApp, verifies completeness, updates your CRM — no human involved.
Custom Websites
Govt Tender Portal
MSME vendor portal with document management, bid tracking, and GeM-ready submission workflow.
Dealer & Distributor Portal
Territory-wise pricing, order management, and inventory visibility for your entire dealer network.
Franchise Network Site
Multi-location website with individual location pages, local SEO, and centralised lead routing.
Export Business Catalogue
Product catalogue with multi-currency pricing, export documentation, and international enquiry forms.
Hospital & Clinic Portal
Appointment booking, doctor profiles, OPD management — patients self-serve, staff focus on care.
Real Estate Portal
Property listings with virtual tours, EMI calculator, and WhatsApp lead capture built in.
School & Institute Portal
Fee payment, online admission forms, and result portal — all in one place for parents and students.
Don't see your use case? Book a free call → — we'll tell you if AI can automate it.
The Platform

Eight layers.
One infrastructure.

We're building this the way AWS built cloud — one primitive at a time, done correctly. Three layers are live. The rest follow the same architecture.

⚡
Execution Layer

Agents take actions across any tool safely. Built-in retry, rollback, failure recovery, and audit logging. Connect once, act everywhere.

Live Beta
🧠
Memory Layer

Persistent, multi-tenant memory across sessions. Agents remember past decisions, build institutional knowledge, and improve over time — per company.

Live Beta
🛡️
Trust & Safety Layer

Configurable permissions per company. Full audit logs. Human-in-the-loop approvals for sensitive actions. Kill switch. This unlocks enterprise.

Partial · Live
👁️
Perception Layer

Real-time event streams so agents sense what's happening across tools automatically — without a human pressing a trigger.

Building · Q3 2026
🔀
Orchestration Layer

Infrastructure for multiple agents to coordinate, delegate, and collaborate. Agent A spins up Agent B. Full visibility into every agent's actions.

Building · Q3 2026
📊
Analytics Layer

Mission control for your agents. Real-time activity feeds, performance metrics, error analysis, and ROI dashboards. Prove value to stakeholders.

Building · Q4 2026
💬
Communications Layer

Smart human-agent communication infrastructure. Right channel, right timing, right format, right person. Two-way — humans respond, agents understand.

Roadmap · 2027
🏪
Agent Marketplace

One-click deploy pre-built agents. Revenue share for builders. Network effects — more agents, more users, more agents. The platform play.

Roadmap · 2027
💡
The AWS analogy is intentional. AWS launched EC2 first — raw compute — because that was the most painful problem. Everything else followed the same architecture. We launched the Execution Layer first. Everything else follows.

This infrastructure
is now available to you.

Everything that powers this agent — execution layer, memory, safety, orchestration — is being packaged as platform API. Request access to use it for your own agents.

Request Early Access →