SahulatAI

🚀 Project Story — SahulatAI Inspiration

Small and micro businesses in Pakistan and emerging markets run their operations manually — responding to customers on WhatsApp, updating inventory in spreadsheets, creating marketing content by hand, and managing bookkeeping with no automation. During the National Agentic AI Hackathon, we spent time talking with local merchants and realized this wasn’t just a niche problem — it was widespread. Most sellers couldn’t afford or learn complicated tools. That’s when the idea of SahulatAI was born: an AI workforce that handles repetitive business tasks for you.

About the Project

SahulatAI is an agentic AI platform that automates sales, customer support, inventory management, marketing, payments, and accounting for small businesses using coordinated AI agents. Businesses onboard with a QR code (especially via WhatsApp), connect their tools like Google Sheets and QuickBooks, and the AI agents start running workflows autonomously. Instead of juggling 5–7 tools, merchants get one AI workforce that saves time and reduces manual errors.

How I Built It

I used a mixture of modern web and AI technologies:

Built with:

Next.js & React (frontend)

FastAPI & Node.js (backend)

AI models (GPT-5.1 / Gemini) for reasoning and agent workflows

WhatsApp Cloud API for conversational automation

Google Sheets API for real-time inventory sync

QuickBooks API for bookkeeping actions

Cloud hosting (Vercel, GCP) for scalable agent orchestration

Each agent is designed to understand context, make decisions, and invoke APIs autonomously. The multi-tenant framework allows multiple businesses to run agents securely and in isolation.

What I Learned

Real users think differently: early prototypes focused on feature depth, but talking to merchants taught us to prioritize ease of use — even if it means simplifying complex workflows.

Onboarding must be frictionless: Traditional dashboards overwhelm micro businesses. A simple QR scan + familiar tools (WhatsApp & sheets) significantly improved adoption.

AI agents need guardrails: Building agents that “do the work” required extensive testing and observability so they don’t make bad decisions — we implemented trace logs and automated error recovery.

Challenges

Integrating real tools like QuickBooks securely while keeping the product simple was technically complex. We built custom MCP servers to manage authenticated access per tenant.

Managing costs around multi-agent AI calls and cloud infrastructure was challenging — each action may trigger many model invocations.

Keeping the system robust when dealing with inconsistent user spreadsheets and messy data required building intelligent data sanitization layers.

Try It Out

👉 Demo: https://sahulatai.app 📹 Video Demo: https://www.youtube.com/watch?v=vww9NZNkNIQ&t=1s