Inspiration
Businesses spend hours on repetitive tasks such as following up with customers, checking invoices, preparing messages, and updating records. Existing AI assistants can generate answers, but they often stop there.
We wanted to build an AI system that does more than respond — one that can understand a business goal, plan the work, execute actions, and verify the result.
That idea led to DocSetuAI: an autonomous AI business operations platform that turns business goals into completed work.
What it does
DocSetuAI lets users give the AI a natural-language business goal.
For example:
"Recover overdue payments from customers whose invoices are more than 7 days overdue."
The AI creates an execution plan and coordinates specialized agents to:
- Find overdue invoices
- Analyze customer information
- Prioritize accounts
- Generate personalized messages
- Request human approval when required
- Execute approved actions
- Schedule follow-ups
- Record activity
- Verify the results
- Generate a final report
Instead of simply telling the user what to do, DocSetuAI works toward completing the task.
How we built it
We built DocSetuAI as a multi-agent application using Gemini and Google Cloud.
The architecture includes an Orchestrator Agent and specialized agents for billing, customers, communication, and follow-ups.
The web application provides the dashboard and task execution interface, while the backend coordinates agent workflows and tools.
Our technology stack includes:
- Gemini
- Google ADK
- Google Cloud
- Cloud Run
- Firestore
- Pub/Sub
- Next.js
- TypeScript
- Python / Node.js
- Docker
Firestore provides persistent application and agent state, while Pub/Sub supports asynchronous workflow execution.
Challenges we ran into
The biggest challenge was going beyond a traditional chatbot.
We needed the AI to actually interact with tools, maintain state, execute multiple steps, and respond to failures.
We also had to design safe human-approval checkpoints for actions such as sending customer communications.
Another challenge was verification. We did not want the system to assume an action succeeded simply because the AI said it did. We therefore designed an explicit verification stage and execution history.
Accomplishments that we're proud of
We are proud that DocSetuAI demonstrates a complete agentic workflow from a simple natural-language goal to a verified business outcome.
The system can:
- Plan multi-step tasks
- Coordinate multiple specialized agents
- Call structured tools
- Maintain persistent state
- Pause for human approval
- Continue execution after approval
- Handle failures
- Verify completed actions
- Provide an auditable execution timeline
Most importantly, the product demonstrates the difference between AI that answers and AI that gets work done.
What we learned
We learned that building reliable AI agents requires much more than a powerful language model.
The quality of tools, structured data, state management, permissions, error handling, human oversight, and verification are equally important.
We also learned that agentic applications work best when the AI is given a clear objective and well-defined tools rather than unrestricted access to the application.
What's next for DocSetuAI
The current version focuses on business operations and overdue-payment recovery, but the architecture can support many more workflows.
Next, we plan to add specialized agents for:
- Sales
- Customer support
- Marketing
- Finance
- Customer success
- Lead qualification
- Internal operations
We also plan to add more integrations, advanced agent memory, organization-level policies, role-based permissions, analytics, and a marketplace for reusable business agents.
Our long-term vision is simple:
Businesses should be able to give AI a goal instead of giving employees another task list.
DocSetuAI — Turn business goals into completed work.
Built With
- agentic-ai
- ai-agents
- artificial-intelligence
- automation
- business-automation
- cloud-run
- docker
- enterprise-ai
- firestore
- gemini
- generative-ai
- google-adk
- google-cloud
- language
- multi-agent-systems
- natural
- next.js
- node.js
- processing
- pub/sub
- python
- saas
- tailwind-css
- typescript
- workflow-automation


Log in or sign up for Devpost to join the conversation.