Submission type
Coded Agent with UiPath SDK
Name
Dhruba Jyoti Kalita
Industry category in which use case would best fit in (Select up to 2 industries)
Compliance
Information technology and services
Complexity level
Advanced
Summary (abstract)
The QA Specialist Agent is an AI-powered coded agent built with the UiPath LangGraph SDK, MCP tools, and context-grounding that automates GitHub pull request reviews.It automates GitHub pull request reviews by fetching PR details, analyzing code using GPT-4o, retrieving QA policies from the ContextGroundingRetriever (“qa-policy-index”), and making data-driven decisions through UiPath MCP orchestration. The agent outputs structured markdown reports, sends automated QA decisions via email, and ensures fast, consistent, and policy-compliant reviews — just like a real QA engineer.
Detailed problem statement
The main challenge this automation addresses is the inefficiency and inconsistency of manual code reviews in modern software development workflows.
Code reviewers often spend significant time analyzing pull requests, ensuring compliance with QA policies, and checking for security or testing gaps — a process that’s slow, subjective, and prone to human oversight.
Additionally, QA engineers struggle to keep reviews aligned with organizational standards and evolving policies, especially in large teams where multiple repositories and contributors are involved.
The QA Specialist Agent solves this by automating the entire code review decision-making process.
Using UiPath’s LangGraph SDK, it fetches live pull request data from GitHub, analyzes the code with GPT-4o, validates compliance through Context Grounding Retriever (qa-policy-index), and leverages MCP tools to deliver structured, auditable QA decisions.
This approach ensures faster reviews, policy-driven accuracy, and real-time insights, reducing manual effort while maintaining high-quality and standardized QA outcomes.
Detailed solution
The QA Specialist Agent is a coded UiPath agent designed to automate the code review process for pull requests in GitHub repositories.
It fits into a developer automation workflow, acting as an intelligent QA reviewer that performs code analysis, policy validation, and automated decision-making — entirely through UiPath’s LangGraph SDK, MCP tools, and context-grounding retriever.
Scope & Purpose
Scope: Automates the review and approval of pull requests (PRs) using real QA policy context and AI reasoning.
Automation Fit: Integrates with a continuous integration/QA pipeline where every PR triggers an autonomous review cycle in UiPath.
Purpose: Reduce manual QA review effort, ensure consistent policy enforcement, and provide explainable decisions (approve/escalate) with supporting rationale.
How It Works
GitHub Pull Request Fetcher
The agent starts when a new PR is raised. It fetches PR details such as repo name, PR number, file diffs, and comments from GitHub.
Findings Extractor (GPT-4o)
Uses GPT-4o to analyze code changes and identify potential QA issues like missing test coverage, hardcoded values, or security risks.
ContextGroundingRetriever (“qa-policy-index”)
Connects to UiPath’s ECS document index to retrieve company QA and compliance policies for context-aware validation.
Decision Maker (LLM + Policy Reasoning)
Combines extracted findings and grounded context to output an Approve or Escalate decision, along with a confidence score and rationale.
Markdown Report Generation
The result is formatted into a markdown report and saved to /artifacts/decision_pr_.md.
MCP SendEmail Tool
Automatically emails the QA Lead or Manager with the decision summary and report link for complete transparency.
UiPath Agent Review Flow
Once all actions are complete, the agent marks the review as Completed in the process trace.
Architecture Overview
Workflow:
GitHub Pull Request → Fetch Latest PR Info → QA Specialist Agent (LangGraph + UiPath SDK)
→ Findings Extractor (GPT-4o)
→ ContextGroundingRetriever (“qa-policy-index”)
→ Decision Maker
→ Markdown Report
→ MCP SendEmail Tool
→ Review Completed
Narrated video link (sample: https://bit.ly/4pvuNEL)
Expected impact of this automation
The QA Specialist Agent delivers measurable improvements in speed, consistency, and compliance for software QA reviews:
-
Time Savings: Reduces manual code review effort by 60–70% by automating policy checks, analysis, and decision reporting.
-
Reduced Repetition: Eliminates repetitive tasks like reading pull request diffs, comparing against QA checklists, and writing review summaries.
-
Improved Compliance: Ensures every review aligns with organizational QA standards through the ContextGroundingRetriever that references live QA policies from UiPath ECS index (qa-policy-index).
-
Autonomous Decisioning: Uses GPT-4o integrated via UiPath’s LLM SDK to make explainable Approve or Escalate decisions — ensuring transparent, data-driven QA validation.
-
End-to-End Automation: With MCP SendEmail, all reports are automatically delivered to QA Leads or Managers, providing full traceability and immediate review visibility.
-
ROI: By reducing manual QA workload and turnaround time per PR from hours to minutes, the agent can save hundreds of engineer-hours per release cycle, directly improving productivity and delivery timelines.
In short, the automation transforms traditional, human-dependent QA review cycles into a fully automated, policy-grounded, and audit-ready process, boosting both developer velocity and compliance assurance across teams
UiPath products used (select up to 4 items)
UiPath AI Center
UiPath Automation Cloud™
UiPath Autopilot™
UiPath Coded Agents
UiPath Integration Service
UiPath Maestro
UiPath Orchestrator
UiPath Robots
UiPath Studio
Automation Applications
Integration with external technologies
Gmail, GitHub
TO-BE workflow/architecture diagram (file size up to 4 MB)
Other resources
Coded Agent Repository : GitHub - DhrubaNits/qa-specialist-langgraph-agent
Materials : QASpecialistCodedAgent - Google Drive
Demo Test Automation Code Repository : GitHub - DhrubaNits/test-automation-demo
Hi there, 
Nov 5–16: Jury evaluation by 
Nov 17: Winners announced 
.
,