Building Software Through Voice A Personal Experiment

Sharing my experience building software through voice conversation with AI teammates - an experimental approach to development.

TLDR

• Built a social services platform MVP using only voice conversation
• Used AI-SDLC framework with MCP-compatible LLMs as teammates
• Completed first draft in 6 weeks, including 197 GitHub issues and 50+ database tables
• Experimental approach combining structured SDLC with AI collaboration
• Demo available at sisimple.vercel.app
• Framework at github.com/mouimet-infinisoft/AISDLC

1. My Journey to This Moment

I'm a software engineer with 25 years of experience, and for the past five years, I've been pursuing an idea: building iBrain, an AI teammate for software development through natural conversation.

The journey hasn't been easy. I've tried thousands of approaches that didn't work, but each failure taught me something valuable. My GitHub contribution graph from 2022 shows this dedication - 3,902 contributions, working almost every day of the year.

2. What I Built

A few years ago, I led a team of five developers in creating a social services management platform. The project lasted three years but, for various reasons, never reached deployment.

Recently, I challenged myself to rebuild this platform solo, applying an experimental approach I had been developing since iBrain: the AI-SDLC framework. Starting from an empty folder and relying solely on voice conversations, I successfully built a complete multi-tenant SaaS platform.

🚀 Key Features Built

An enterprise-grade case and contact management platform designed for secure, multi-tenant environments with automation, scheduling, and reporting capabilities.

🏢 Enterprise-Grade Infrastructure

• Multi-tenant architecture with full data isolation
• Role-based access control and permissions
• Secure authentication and tenant administration
• Multi-language support

👥 User & Organization Management

• Organization and tenant management
• User account management and authentication
• User profiles and settings
• Employee directory and management

📁 Core Platform Functionality

• Case file management with full lifecycle tracking
• Contact management and relationship tracking
• Request intake and tracking system
• Document management with versioning and access controls
• Observation notes and case annotations
• Scheduling and calendar system

⚙️ Workflow & Automation

• Workflow automation engine
• Template-based report generation
• Role-specific dashboards and task views

📊 Insights & Reporting

• Customizable reporting and analytics

🎥 Watch It in Action

See how the voice-driven development process works and explore the resulting platform in this demonstration:

This video demonstrates the voice-driven development process, showing how AI teammates collaborate to build enterprise software. You'll see real examples of voice-based problem solving, AI-assisted implementation, and the resulting platform features.

📸 Feature Highlights

🛠 Workflow Automation Dashboard

Shows an automated request processing pipeline with real-time status tracking for operational efficiency.

📅 Calendar with Resource Management

Visualizes scheduled tasks, detects conflicts, and optimizes staff resource allocation.

👥 Staff Scheduling Matrix

Provides a clear overview of staff availability to simplify shift planning and workforce coordination.

🚀 Try It Yourself

See the result in action:

Demo: sisimple.vercel.app

• Email: robert.gagnon@infinisoft.world
• Password: Password123!

3. The Development Process

Through voice conversation, the AI teammates:

1. Created complete project documentation and planning
2. Scaffolded the entire project structure
3. Broke down work into GitHub issues with proper hierarchy
4. Implemented features with continuous review
5. Generated comprehensive technical documentation

📌 GitHub Project Board Integration

Visual project management using GitHub milestones and issue hierarchy to track development progress and roadmap alignment.

🧱 Hierarchical Issue Management

Supports a three-level issue structure with clear parent-child relationships, enhancing task traceability and accountability.

🗂 Case Management Timeline

Dedicated interface for viewing case progression, attaching documentation, and managing case notes across the lifecycle.

My role was primarily supervision and direction - the AI teammates handled the implementation details while I focused on strategic decisions.

What makes this particularly interesting is that while it took 6 weeks, much of that time was spent developing the AI-SDLC framework itself. With the framework now established, similar projects could potentially be completed much faster.

4. Results & Metrics

Starting from an empty folder, through voice conversations with AI teammates, we created:

Project Structure & Documentation:

• 197 GitHub issues organized in a three-level hierarchy
• 14 development milestones with clear progression
• Complete technical specifications and architecture documents
• API documentation and implementation guides

Technical Implementation:

• 50+ database tables with proper normalization and relationships
• 100+ React components following consistent patterns
• 40+ database migrations with rollback capability
• Full authentication and authorization system
• Multi-tenant data segregation with row-level security
• Template engine for document generation
• Workflow automation system

Quality Assurance:

• Test coverage across core functionality
• Basic CI/CD pipeline setup
• Security policy implementation

All of this was achieved through natural conversation - I supervised and made strategic decisions while the AI teammates handled implementation details and documentation.

The resulting first draft MVP includes all core features needed for a production system, though it would need significant iteration and refinement before actual deployment.

5. The Experimental Framework

The AI-SDLC framework combines structured development practices with AI collaboration through the Model Context Protocol (MCP). Here's how it works:

🤖 Meet Your AI Team

Business Analysis Requirements gathering and business case creation	System Architecture Technical design and system specifications	Project Management Workflow coordination and project structure
Functional Analysis Detailed functional requirements and specifications	Development Leadership Implementation planning and technical strategy	Development Code implementation and testing
Frontend Development UI/UX implementation and responsive design	Quality Assurance Testing, code review, and quality validation	DevOps & Deployment Infrastructure, CI/CD, and deployment automation

AI-SDLC Workflow

The development process follows two main phases:

Phase 1: Strategic Planning & Design

Phase 2: Implementation & Delivery

How We Collaborate

The framework connects these AI teammates through an MCP-compatible application, allowing natural voice discussions about development tasks. I tested this successfully with GitHub Copilot and Augment Code, though it should work with any LLM that supports MCP.

Each conversation follows a structured workflow:

1. Discuss requirements and objectives with Sarah and Taylor
2. Plan technical approach with Alex and Casey
3. Implement solutions with Mike and Riley
4. Review and validate with Sam

The AI teammates maintain context throughout the development process while following standard SDLC practices.

Components used:

• Standard tools: VS Code, GitHub
• AI Platforms: GitHub Copilot, Augment Code
• Custom tools: Voice MCP, Diagram MCP, N8N MCP
• Tech stack: Next.js 15, React 19, Supabase, PostgreSQL

6. Voice-Based Discussion Approach

Instead of typing commands or code, I expressed ideas through natural conversation. The AI teammates:

• Asked clarifying questions
• Proposed solutions
• Implemented features
• Created pull requests
• Maintained code quality

This wasn't about voice commands - it was collaborative problem-solving through conversation.

7. Key Differences

Traditional development involves significant coordination overhead between requirements, implementation, and validation.

This experimental approach:

• Reduced communication overhead
• Enabled rapid iteration
• Maintained engineering standards
• Preserved project context
• Generated documentation automatically

8. Challenges & Solutions

What worked well:

• Project management
• Complex algorithm implementation
• Database schema design
• Architecture consistency
• Documentation generation

What needed iteration:

• UI refinement
• Edge case handling

Specific challenges we solved:

• Multi-tenant data isolation
• Resource scheduling conflicts
• Workflow state management
• Document generation templates

9. Conclusion

What started in 2018 as a spark of curiosity has become an ongoing vision.

I didn’t invent LLMs or make a scientific breakthrough—but I asked myself: “What if we treated AI like collaborators, not just tools?”

That simple idea has guided years of building, failing, and reimagining. It’s been my creative sandbox: experimenting to give AI personality, awareness, and a sense of unique collaboration.

It’s far from polished. Not commercial. But it’s finally alive—and deeply personal.

If any part of this resonated with you—or sparked a thought—I’d genuinely love to hear it. I’m also considering a deeper technical follow-up—would that interest you?

No pressure. No commitments. Just curiosity and conversation.

Project Links

• Framework: github.com/mouimet-infinisoft/AISDLC
• Project Board: Si Simple Platform Development
• Repository: github.com/mouimet-infinisoft/sisimple
• Demo: sisimple.vercel.app
• Youtube: Demo Video

Note: This is experimental work, shared to encourage discussion and exploration.