AI-Assisted Development

Our Philosophy: Experience-Driven AI Development

UnaMentis is built with AI-assisted development from the ground up, but this is not "vibecoding." The project founder brings over 30 years of experience in technology, the majority spent as a developer, with years contributing to open source projects. Every architectural decision, every tool selection, and every quality standard is informed by decades of real-world software engineering experience.

The goal is ambitious: use AI not just to move faster, but to approach the quality and review standards achieved by a thoughtful, attentive human developer. We have deep respect for that standard. A skilled developer bringing full attention to code review, architecture decisions, and quality assurance is not easily replicated by any single tool. However, when six or seven layers of AI-driven tools and processes work together, each with overlapping review and complementary perspectives, the cumulative effect can begin to approximate that level of rigor. We are documenting an ongoing experiment in what becomes possible when deep experience guides this layered approach with intention.

AI handles the repetitive, error-prone aspects of software development while human experience guides architecture, quality standards, and the creative problem-solving that makes UnaMentis unique. The combination enables a small team to build and maintain a sophisticated, multi-platform voice AI tutoring system.

AI Tools We Use

Our AI-assisted development workflow combines multiple specialized tools, each chosen for its strength in a specific domain. Together, they form a comprehensive system that touches every aspect of our development process.

The Code Quality Initiative

To achieve enterprise-grade quality with a small team, we implemented a systematic 5-phase Code Quality Initiative. Each phase builds on the previous, creating layers of automated protection that catch issues progressively earlier in the development cycle.

Key Achievements

Phase 1: Foundation

The foundation phase automates existing manual quality gates across iOS, Server, and Web components. The goal: make quality enforcement invisible and unavoidable.

Pre-Commit Hooks

Pre-commit hooks are automated checkpoints that run every time a developer tries to save code changes. Think of them like a spell-checker that runs automatically before you can send an email. If the code has formatting problems, style violations, or accidentally included passwords, the save is blocked until those issues are fixed. This catches problems at the earliest possible moment, when they are easiest and cheapest to fix.

Our unified hook system runs automatically before every commit, completing in under 30 seconds while checking code across all three platforms:

Hook Bypass Auditing

Developers can skip pre-commit checks in emergencies using a special flag. While sometimes necessary, frequent bypasses indicate a problem. Our audit system tracks every bypass, creating visibility into whether quality gates are being circumvented and enabling conversations about why.

Dependency Automation (Renovate)

Manual dependency tracking is eliminated. Renovate handles everything automatically:

• Schedule: Updates run Mondays before 6am, minimizing disruption
• Grouping: iOS, Python, and npm dependencies grouped separately for focused review
• Auto-merge: Security patches, patch updates, and dev dependencies merge automatically
• Manual review: Major version updates and breaking changes require human approval

Coverage Enforcement

Code coverage measures how much of our code is actually tested by our automated tests. When we say "80% coverage," it means that when all our tests run, they exercise at least 80% of the code paths in the application. The remaining 20% represents code that is not directly tested, which could hide undetected bugs.

We treat coverage as a hard gate, not a suggestion. If the iOS codebase drops below 80% coverage, the build automatically fails and cannot proceed. This forces new code to include tests and prevents gradual erosion of test quality over time. Coverage is extracted automatically from Xcode test results, so enforcement is completely automated.

Phase 2: Enhanced Quality Gates

Phase 2 extends quality enforcement with nightly testing, performance regression detection, and comprehensive security scanning.

Nightly End-to-End Testing

Every night at 2am UTC, comprehensive end-to-end tests run against the full system:

• iOS E2E tests with real API keys (from GitHub Secrets)
• Latency regression tests using the provider comparison suite
• Full voice pipeline validation
• Automatic GitHub issue creation on failure with "nightly-failure" label

Performance Regression Detection

Voice applications live and die by latency. Our latency test harness ensures we never ship a slower release:

Multi-Layer Security Scanning

Security is not a single check. It is a layered defense where multiple specialized tools each look for different types of problems. No single tool catches everything, but together they provide comprehensive protection:

Phase 3: Feature Flag System

Feature flags are on/off switches in code that control whether a feature is visible to users. They let us deploy new code to production but keep it hidden until we are ready to reveal it. This is powerful for several reasons:

• Safe rollouts: Enable a new feature for 1% of users first, watch for problems, then gradually expand to everyone.
• Instant rollback: If a new feature causes problems, disable it with a configuration change instead of deploying a code fix.
• A/B testing: Show different experiences to different users and measure which performs better.
• Operational control: Disable expensive features during high traffic periods or outages.

Self-Hosted Unleash Infrastructure

We run our own feature flag system using Unleash, an open-source platform that commercial services like LaunchDarkly charge $75+ per month for. Self-hosting gives us full control over our data and zero ongoing costs:

Flag Lifecycle Management

Feature flags have a lifecycle. A flag created for a gradual rollout should be removed once the feature is fully launched. Forgotten flags accumulate as "flag debt," making code harder to understand and maintain. Our automated audit system prevents this:

• Ownership: Every flag has a designated owner and a target removal date set when the flag is created.
• Expiration enforcement: Weekly automated scans detect flags older than 90 days and create cleanup tickets.
• Early warning: CI automatically creates GitHub issues when flags approach their expiration date.
• Visibility: Any code change that adds, modifies, or removes a flag triggers an automatic comment on the pull request for review.

Phase 4: Observability & DORA Metrics

You cannot improve what you do not measure. Phase 4 provides visibility into quality trends and engineering health through industry-standard DORA metrics.

DORA Metrics (Apache DevLake)

DORA (DevOps Research and Assessment) is the largest and longest-running research program studying what makes software teams effective. The research, published in the book Accelerate by Dr. Nicole Forsgren, Jez Humble, and Gene Kim, identified four key metrics that distinguish elite engineering teams from the rest. We track all four:

Research shows teams that excel at these metrics are twice as likely to exceed their organizational performance goals and report significantly higher customer satisfaction. We use Apache DevLake, an open-source platform, to collect and visualize these metrics automatically from our GitHub repositories.

Quality Dashboard

Daily automated metrics collection provides ongoing visibility:

• CI/CD success rates across iOS, Server, and Web pipelines
• Pull request metrics (count, average size, review time)
• Bug metrics (open count, closed in 30 days, age distribution)
• Trend analysis with 90-day retention for pattern detection

Phase 5: AI-Powered Code Review

The final phase brings AI into the review process. Every pull request receives automated analysis from CodeRabbit, configured for maximum issue detection with language-specific rules for our tech stack.

Review Configuration

CodeRabbit is configured in "assertive" mode for comprehensive coverage:

Results & The Road Ahead

The Code Quality Initiative is an ongoing journey. All five phases are now substantially complete, with continuous refinement ongoing. Here is where we stand:

Current Quality Gates

Implemented Advanced Features

Planned Advanced Features