PROJECT

AI automation for connected business workflows

Custom software development, AI agent design, Workflow automation, API and webhook integration, CRM integration, Document and knowledge processing, Human approval workflows, Workflow documentation

  • Industry: Business services
  • Latvia Location: Latvia
  • Delivery time: 16 weeks
  • Technologies: JavaScript / TypeScript (Node.js), Python
4+ months

This client spends time connecting AI agents, client data, APIs, and business software for production workflows. The solution is AI automation software for the agency: it takes in business data, messages, and files. It gives back workflow decisions, structured results, and completed tasks for the agency’s business clients, helping reduce manual work.

Summary

Problem

The agency spends time joining AI agents, business systems, data, and manual approval steps across separate tools.

Solution

The solution became a modular platform for building agents, connecting services, processing information, and monitoring multistep workflows.

Result

The platform is made to reduce manual work and rework from incoming business data to a completed task.

Client story

The client is an AI automation agency from Latvia. It creates production workflows for business clients and connects their CRM, sales, and support work.

The agency works with APIs, webhooks, databases, and business applications. Manual steps between systems cost time, especially when a workflow needs a decision, a write-enabled update, or human review.

The request was to build AI agents and multistep workflows that perform business tasks, use external tools, make decisions, and reduce manual intervention. The agency also needed grounded responses based on client knowledge and connected business data.

The solution was shaped in separate modules. Workflow design, execution monitoring, and AI review each received a clear place.

Approval points, failure handling, data mapping, and handover materials kept each client workflow clear.

The client asked for

  1. Build AI agents and workflows for business tasks and decisions
  2. Connect APIs, webhooks, CRMs, databases, and business applications
  3. Process documents, messages, files, and knowledge bases
  4. Use LLMs, RAG, vector databases, and knowledge for grounded results

Why ready-made software did not fit

Ready-made products cover parts of the task: agent integration tools connect AI to business software, while enterprise automation platforms handle legacy applications. Other products focus on marketing agency management or specialist marketplaces. The required solution combines business systems, AI agents, multistep decisions, monitoring, human review, and handover documentation.

Project overview

The solution was designed around a visual workflow builder where the agency creates AI agents, adds business rules, and connects tools. API inputs and approval points stay visible in the same design.

To handle running work, a monitoring module was provided for execution history, workflow status, and alerts. Logs and recovery controls give the agency a clear place to inspect errors.

The AI assistant connects user requests with retrieved knowledge and produces structured outputs. Sensitive or uncertain cases move to a manual-review queue, while approved actions continue through connected systems.

Special attention was paid to integration work. APIs, webhooks, and CRM records are mapped to workflow steps.

Database records and JSON fields support data exchange, with write-action previews before information moves back into a business platform.

The solution also includes workflow diagrams and architecture descriptions. Setup guidance, maintenance instructions, troubleshooting procedures, and prompt materials help the agency operate and hand over each workflow.

Key features delivered

  • Visual AI workflow builder
  • AI agent configuration
  • Tool and API connections
  • Decision and human approval steps
  • Workflow execution history
  • Failure alerts and recovery controls
  • AI assistant with retrieved knowledge
  • CRM and database data mapping
  • Write-action preview
  • Workflow documentation and handover

Can be added

+ Customer-facing chatbots + Voice agents + Messaging automations + Lead qualification flows + Appointment booking + Outreach and follow-up automation + Additional client workflow modules

Who else can use this solution

These businesses have a similar task, and the same solution can be adapted to connect their data, tools, AI agents, and operational steps.

A business services company

Task It connects sales, support, finance, HR, or project management tasks across separate platforms.
How the solution helps The solution joins those systems into monitored workflows with AI decisions, structured results, and approval steps.
Business services

An organization building an AI-powered SaaS platform

Task It needs agents, business data, APIs, and knowledge sources to work together inside its product.
How the solution helps The solution provides workflow design, integrations, grounded responses, execution monitoring, and documentation.

A business automating customer communication

Task It processes customer messages, lead information, scheduling requests, and follow-up actions.
How the solution helps The solution connects communication data to AI assistance, CRM updates, routing, and human review.

Technology stack

JavaScript / TypeScript (Node.js)

Application and integrations

This language supports the workflow interface, agent connections, webhooks, and API integrations.

Python

Agents and data processing

This language supports AI workflow logic, data handling, CRM connections, and documentation tools.

PHP

Business system connectors

This language supports selected CRM, webhook, Google, and workflow service connections.

@graph8/sdk

CRM and sales actions

This SDK connects CRM, enrichment, sequences, campaigns, and B2B sales tracking.

dynamics365crm-python

Dynamics CRM integration

This Python API wrapper exchanges records with Dynamics 365 CRM.

zohocrmsdk8-0

Zoho CRM integration

This SDK connects workflow steps with Zoho CRM version 8 APIs.

svix

Webhook delivery

This library sends and verifies webhook events between connected services.

django-anymail

Email and webhooks

This library connects email backends and webhook events from supported email services.

n8nc

N8n workflow control

This command-line tool synchronizes and manages n8n workflows.

JSON

Structured data exchange

JSON carries mapped fields, workflow inputs, outputs, and system responses.

Integrations

CRM systems

The solution reads CRM records and sends write-enabled updates through connected CRM APIs.

Webhooks

Webhooks bring events and workflow inputs into the automation platform and return workflow events to other services.

Databases

Database records provide workflow data and receive structured outputs where configured.

Google Drive and Google Sheets

The solution exchanges documents and spreadsheet data with Google Workspace services.

Email services

Email connections bring messages and attachments into workflows.

n8n

The solution triggers, synchronizes, and tracks n8n workflows through API and webhook connections.

The result

Design workflows with agents and approvals

The workflow builder gives the agency one place to shape AI agents and their multistep actions.

The canvas also supports editing as a client workflow changes, while the agent configuration keeps task instructions and workflow actions clear.

How it works

  1. The workflow canvas shows AI agent steps, actions, decisions, and approval steps.
  2. The agent configuration contains task instructions and business rules.
  3. Tool and API controls connect inputs and workflow actions.
  • Create and edit AI automation workflows
  • Add approval steps for uncertain or sensitive decisions
Tip: Approval points keep selected decisions with the client team when automation alone is not appropriate.
Workflow builder and agent settings

Workflow builder and agent settings

Functions on this screen
Design AI automation workflows Create AI Agents Backend API Integration Edit workflows

See failures before they interrupt work

The monitoring screen gives the agency a clear view of running workflows, completed tasks, failures, and alerts. Execution history and logs show the information needed to investigate a problem.

Failure details and recovery controls give the agency one place to inspect a workflow problem.

How it works

  1. The execution list shows workflow inputs, statuses, completed tasks, and decisions.
  2. The execution log shows logs, error reports, and failure alerts.
  3. The recovery control provides retry or recovery actions for failed executions.
  • Review execution history
  • Inspect failure details and alerts
  • Track workflow decisions
Note: The monitor keeps failure information next to the workflow execution that produced it.
Workflow execution monitor

Workflow execution monitor

Functions on this screen
Monitor workflows for errors

Review grounded answers and sensitive actions

The AI assistant combines retrieved knowledge with user requests and returns responses or structured results. The review queue separates cases that need a person’s decision from routine workflow results.

This gives the agency a controlled way to handle support tasks, client knowledge, and actions that require approval.

How it works

  1. The conversation panel shows user requests and AI responses.
  2. The retrieved knowledge area shows knowledge-base content used for the response.
  3. The review queue shows approval requests with approve or reject controls.
  • Preview structured outputs
  • Route selected cases to manual review
AI assistant and review queue

AI assistant and review queue

Functions on this screen
AI Integration Automate support tasks

Map data between business systems

The integration screen connects APIs, webhooks, CRMs, and databases to workflow steps. A mapping table makes incoming fields and workflow outputs easier to understand before an action updates a connected system.

The screen also shows the data that moves between workflow steps and connected systems.

How it works

  1. The connected systems list shows API, webhook, CRM, and database connections.
  2. The data mapping table maps JSON fields, CRM records, and database records to workflow steps.
  3. The integration screen shows workflow outputs and connected system data.
  • Map incoming business data
  • Review workflow outputs
Integration and data mapping

Integration and data mapping

Functions on this screen
Backend API Integration Integrate business systems

Keep operations understandable after handover

The documentation screen turns the finished workflow into an operating reference. It brings the workflow diagram, architecture description, and setup guide together.

Maintenance instructions, troubleshooting procedures, and a prompt and agent instruction library help the agency team maintain the logic behind each workflow.

How it works

  1. The workflow diagram shows workflow steps and connected systems.
  2. The architecture description and setup guide show configuration values and operating instructions.
  3. The maintenance and troubleshooting instructions show failure procedures and agent instructions.
  • Document workflow architecture
  • Keep prompts and agent instructions together
Workflow documentation and handover

Workflow documentation and handover

Functions on this screen
Document workflows and training

What the solution gives

With this solution, the agency can:

  • Build AI agents for specific business tasks
  • Automate multistep workflows across connected systems
  • Process documents, messages, files, and knowledge-base content
  • Route sensitive decisions to human approval
  • Monitor workflow failures, alerts, logs, and recovery actions
  • Hand over documented workflows with setup and troubleshooting guidance

Could ready-made systems do the job?

Before the design, the task was compared with ready-made products by their own descriptions.

What the client needs Agent integration tools AI agent integration tools Enterprise automation platforms StitchOps Agency operating systems Markezy Custom solution
Build agents and multistep workflows partly —
Connect APIs, CRMs, and databases —
Process files and knowledge partly partly —
Ground results with client knowledge partly — —
Fits the client's own rules: stages, roles, reports partly partly —
All steps in one place — — —
No manual steps between programs — — —

By the public descriptions of the products and their category.

Ready-made products cover important parts of agent integration, enterprise automation, and agency operations. A ready-made product is enough when one of those parts is all that the business needs.

This client needed its own software because the required agents, connected systems, grounded knowledge, monitoring, human review, documentation, and client-specific rules were not covered together in one place.

In the end

In the end, the solution was made to take manual work out of the path from a client request and business data to a documented, monitored workflow.

FAQ

Thinking about building a similar system? Read the answers below to see how AI agents, integrations, data processing, and human review fit into one automation platform.

Need something similar? Let's talk
Or browse more projects