Let's talk

Founder's Journey

Building an AI Company from a Small Town in India

You don't need a metro city to build world-class AI. From Palghar, I lead a studio of three humans and fifteen AI agents to ship complex agentic systems.

By Divyaprakash JhaUpdated 4 min readTopic: Founder

Photo for Building an AI Company from a Small Town in India

Building a global AI company is possible from small-town India by using agentic architecture and a high human-to-agent ratio. My studio operates with just three human operators and a fleet of fifteen AI agents to ship complex systems from Palghar, Maharashtra.

The Reality of Building AI in Palghar

You do not need a metro city to build world-class technology. From my base in Palghar, I lead a specialized studio that leverages the power of autonomous agents to compete on a global scale.

  • Agentic Architecture: We use specialized AI agents to handle heavy lifting, allowing a tiny team to produce the output of a large organization.
  • The Studio Model: My current setup consists of three human operators directing fifteen specialized AI agents.
  • Infrastructure Independence: We run our own GPU and CPU servers with open models to ensure we are not entirely dependent on paid third-party APIs for bulk work.
  • Global Reach: We build products like GodEngine and AskShiva that solve complex problems for users worldwide from a local Indian base.

The Human-Agent Ratio: Scaling Small

The traditional agency model relies on hiring dozens of people to handle different tasks. I have replaced that overhead with a "Human-Agent Ratio" that prioritizes leverage over headcount.

In our studio, humans focus on high-level architecture and intent—deciding what should be built and why. Once the intent is set, the agents handle the execution in parallel. This allows us to ship complex software at a speed that would overwhelm a much larger, traditional team.

To reflect the specialized nature of these digital workers, we name our agents using Sanskrit or Hindi terms to denote their specific craft:

  • Vastu: Architecture and structure.
  • Sutra: Code and logic.
  • Akshar: Lettering and text.
  • Vani: Words and communication.
  • Ganit: Numbers and mathematics.
  • Parakh: Testing and verification.

This approach turns a small team into a high-output engine of production.

From 500 Trading Systems to Agentic AI

My journey into AI was not accidental; it was an evolution of a decade spent in the unforgiving markets of Gold and FX. Starting in 2016, I learned that markets do not care about your confidence—they only care if your system holds up when conditions change.

During those ten years, I built and shipped more than 500 Expert Advisors (EAs) for MetaTrader 5. These were essentially "small agents": automated programs that read signals, make decisions, and manage risk without a human clicking buttons. I served around 100 clients through my studios, CandleForge and WICKFURY, teaching me the discipline of building automated systems.

When Large Language Models (LLMs) arrived, the transition felt natural. The core questions of trading—What does the system know? What is it allowed to do? How do you check it before it costs you money?—are the exact same questions we answer when building agentic AI. I moved from building automated trading tools to building decision intelligence engines.

Building for Real-World Utility

Many AI startups focus on "clever" technology that lacks practical application. We focus on the gap between a "clever answer" and "finished work."

While most chatbots provide a paragraph of text, we build tools that complete tasks. For example, a small business owner in India doesn't just want a paragraph explaining how to make an invoice; they want the actual invoice, formatted correctly, sent via WhatsApp in their local language.

This philosophy drives our two primary products:

  • AskShiva: An assistant designed for the Indian context. It handles tasks like generating GST bills, translating messages into Marathi or Hindi, and managing follow-ups on WhatsApp. It is built to handle the reality of Indian business: WhatsApp, GST, and UPI.
  • GodEngine: A decision intelligence engine. Instead of just chatting, it treats a question as a model of the world, showing the reasoning, causal links, and sources behind a recommendation.

Technical Independence and Infrastructure

To maintain control over our costs and reliability, we do not rely solely on third-party API providers. We invest in our own hardware, running our own GPU and CPU servers. This allows us to run open models for bulk AI work, ensuring that a single provider outage doesn't paralyze our production.

Our technical architecture is built on a "plan, act, verify, and remember" loop. This is critical for reliability in business settings. We don't just ask an AI to do something; we implement a pipeline where the AI plans the steps, executes them using tools, verifies the result, and then logs the outcome into its memory.

Feature Standard Chatbots Our Agentic Architecture
Output Fluent text/sentences Finished work (files, code, invoices)
Reasoning Hidden in the prose Auditable, traceable, and structured
Reliability Prone to "hallucinations" Verified through "plan-act-verify" loops
Independence Total reliance on APIs Hybrid (APIs + local GPU/CPU servers)

The Future: BusinessOS

The ultimate goal of our work is to solve the fragmentation that plagues most Small and Medium Enterprises (SMEs). Most businesses run on a mess of disconnected tools: leads in a spreadsheet, customers on WhatsApp, HR in one app, and invoices in an email. Nothing talks to each other, and no one has a single "source of truth."

We are building BusinessOS to be the AI-native operating system that fixes this. It is not just another tool bolted onto a stack; it is a unified foundation. It provides:

  • Identity & Permissions: A single engine for roles and access control.
  • Data Isolation: Secure, database-level separation for different organizations.
  • Auditability: An append-only, searchable audit log of every action taken.
  • Unified Foundation: One search, one file store, and one event bus that connects every module.

We are moving from simple chat to a complete, integrated business operating system. This is where the power of Shiva Agents meets the structural needs of a modern company, creating an environment where AI agents can work as an "always-on" company alongside humans.

Questions people ask

Can you build a global AI startup from a small town in India?

Yes. By using agentic architecture and a small team of humans supported by a fleet of AI agents, you can ship high-quality products from anywhere, such as Palghar.

Do you need a large team to build an AI company?

No. I run a studio with just three human operators and fifteen AI agents, using parallel agentic workflows to achieve high output.

Why use open models instead of just paid APIs?

Running our own GPU and CPU servers with open models allows us to handle bulk AI work without being entirely dependent on paid API providers.

What is the advantage of agentic AI over standard chatbots?

Standard chatbots provide fluent text, but agentic AI can plan, use tools, verify its own work, and complete multi-step tasks like generating a GST invoice.

How do you ensure AI agents are reliable for business?

We use a 'plan, act, verify' loop and implement strict autonomy levels and approval gates for any high-risk actions.