EC

Erick Castillo

Lead .Net/Azure Developer

View Portfolio

Hexaware Technologies

Lead .NET Developer

October 2025 - December 2025

The Problem

The five-person ELA development team was shipping .NET features at a traditional pace — handwritten CQRS handlers, manually scaffolded EF Core configurations, and inconsistent test coverage across modules. AI tools were available to the organization but used inconsistently: some developers leaned on Copilot for autocomplete, others ignored it entirely, and no one had a shared convention for prompts, context, or AI-assisted code review. The result was uneven velocity, duplicated effort across the team, and no defensible way to measure whether AI tooling was actually helping. On a federal contract with hard delivery milestones, the team needed a repeatable workflow rather than ad-hoc AI use.

The Solution

Defined and rolled out team-wide AI-native engineering standards covering spec-first development, prompt and context conventions, and AI-augmented code review checkpoints. Integrated Claude Code, GitHub Copilot, and M365 Copilot into the daily workflow with clear guardrails on what each tool was best for. Established markdown-based Technical Design Documents as the single source of truth that AI agents would consume to scaffold new modules, then trained the team on the workflow through mob-coding sessions and code reviews.

Implementation

Wrote the team standards doc covering: TDD structure (problem, constraints, architecture, acceptance criteria), prompt conventions for scaffolding .NET projects, context-file patterns for keeping AI agents grounded in the ELA codebase, and review checkpoints where AI-generated code had to pass before merge. Ran mob-coding exercises where the team would author a TDD together, then watch Claude Code scaffold the CQRS handlers, EF Core configurations, and xUnit tests live — turning AI tooling into a shared practice rather than a private habit. Built CLAUDE.md-style context files at the repo root so every developer's agent had the same architectural ground truth (Clean Architecture conventions, naming, project layout). Defined an AI code review checkpoint in the GitLab merge-request template so reviewers knew when AI-generated code needed extra scrutiny (data access, auth, anything touching FEMA PII). Mentored the team on prompt design, when to trust AI output, and when to throw it away and write it by hand.

Technologies

Claude Code (Anthropic) GitHub Copilot M365 Copilot ASP.NET Core .NET 9 C# CQRS / MediatR Entity Framework Core xUnit GitLab Markdown TDDs Spec-First Development Context Engineering MCP servers

Impact & Results

Shifted the team from ad-hoc AI use to a defined, repeatable workflow with shared conventions and review gates. Time-to-first-working-build for new modules dropped meaningfully once TDD-driven scaffolding became the default — handlers, EF Core configs, and tests generated from spec instead of typed by hand. Code review quality improved because reviewers had a clear checklist for AI-generated code, and onboarding got faster because the standards doc and CLAUDE.md context files gave new developers a runway. The team built a defensible story for leadership that AI tooling wasn't just autocomplete — it was a productivity multiplier with measurable gates.

Time-to-first-working-build reduced ~60% via TDD + AI scaffolding

5-developer team standardized on a single AI-native workflow

Markdown TDDs adopted as source of truth for new modules

AI code review checkpoint added to every merge request

CLAUDE.md context files deployed across repos for consistent agent grounding

Mob-coding cadence established for shared skill development

This entry is public
0 Likes
0 Vouches

Vouches (0)

Vouches verify the authenticity of this work. Each voucher explains their connection to the work.

Loading vouches...