AI .NET React Azure DevOps

Sprint Planning AI Agent

Autonomous AI agent for Agile sprint planning using Claude's ReAct pattern

Live Demo →

Overview

An autonomous AI agent that automates Agile sprint planning by analyzing Azure DevOps backlogs, calculating team velocity, and using Claude’s Sonnet 4 model to create balanced sprint plans.

The Challenge

Engineering teams spend 2-4 hours in sprint planning sessions manually reviewing backlogs, calculating capacity, and balancing work across team members. This process is repetitive, time-consuming, and pulls the entire team away from development work.

The Solution

A production-ready AI agent that:

  • Connects directly to Azure DevOps via REST API
  • Analyzes backlog items (stories, bugs, tasks) with full context
  • Calculates team velocity from historical sprint data
  • Autonomously plans sprints using the ReAct (Reasoning + Acting) pattern
  • Explains every decision with transparent reasoning

Additional AI Capabilities

The system also includes an Agile Analyzer API with:

  • Code Review Analysis - Identifies issues, security concerns, and best practices
  • Retrospective Analysis - Extracts themes, sentiment, and actionable items
  • Tech Debt Prioritization - Ranks technical debt by ROI and business impact

Technical Architecture

Backend APIs (.NET 10)

  • Sprint Planning Agent - Orchestrates AI-powered sprint planning
  • Agile Analyzer - Provides AI analysis for code reviews, retros, and tech debt
  • Comprehensive test coverage (109 NUnit tests across two test projects)
  • Tiered rate limiting (3–10 req/min depending on endpoint)
  • Swagger documentation

Frontend (React + TypeScript)

  • Modern, responsive UI with IdeaRoost branding
  • Real-time API communication
  • Vite build system
  • TypeScript for type safety

Infrastructure & DevOps

  • Azure App Service (Free tier F1)
  • Azure Static Web Apps (Free tier)
  • GitHub Actions CI/CD - Automated build, test, and deployment
  • Secure configuration - All secrets managed via GitHub Secrets
  • Custom domain with SSL

Key Features

ReAct Pattern Implementation

The agent uses Claude’s ReAct (Reasoning + Acting) pattern:

  1. Reason - Analyzes the current state and decides what to do
  2. Act - Calls appropriate tools (get backlog, get velocity, create sprint)
  3. Observe - Reviews results and continues until task is complete

This gives Claude genuine autonomy - it chooses which tools to use based on context, not a hardcoded workflow.

Tool Use API

Three autonomous tools:

  • GetBacklogItems - Retrieves work items from Azure DevOps
  • GetTeamVelocity - Calculates sprint velocity from history
  • CreateSprint - Creates and populates sprints in Azure DevOps

Cost Management

  • Tiered rate limiting prevents runaway costs (3–10 req/min depending on endpoint)
  • ~$5-15/month total operating cost

Tech Stack

Backend:

  • .NET 10
  • ASP.NET Core Web API
  • Azure DevOps REST API
  • Claude Sonnet 4 (Anthropic API)
  • NUnit for testing

Frontend:

  • React 19
  • TypeScript
  • Vite

Infrastructure:

  • Azure App Service
  • Azure Static Web Apps
  • GitHub Actions
  • Azure CLI

Results

  • ✅ Production-ready - Full error handling, rate limiting, security
  • ✅ Fully automated CI/CD - Push to deploy
  • ✅ Cost-effective - Runs on Azure free tier
  • ✅ Secure - All secrets in GitHub, no credentials in code
  • ✅ Professional - Comprehensive tests, Swagger docs, CORS configured

Live Demo

Frontend: devteamaiassistant.idearoost.com

What I Learned

  1. Agentic AI is production-ready - The ReAct pattern works reliably
  2. Azure free tier is powerful - Can handle real production workloads
  3. Rate limiting is essential - Prevents cost overruns with LLM APIs
  4. CI/CD saves hours - Automated deployment removes deployment friction
  5. Tool use > Prompting - Structured tool interfaces beat prompt engineering

Interested in AI-augmented development or .NET/Azure consulting? Let’s talk

← Back to portfolio