Highlights
Industry:
Digital Entertainment and Mobile Products
Project scope:
- AI agentic platform for software development, QA automation, business analytics, marketing creative analysis, and IT administration
The Client
A digital entertainment company with several high-load products, international users, and multiple engineering, QA, analytics, marketing, and IT teams.
The Project
The client wanted to reduce manual work across several internal teams.
Development teams needed a faster way to turn Jira tasks into reliable pull requests. QA teams wanted to increase automated test coverage without spending too much time on repetitive test creation and maintenance.
Business teams needed easier access to analytics without depending on data specialists for every request. Marketing teams wanted to organize and analyze large volumes of short ad videos. IT administrators needed help with routine internal requests coming from different channels.
The goal was to build a practical AI agentic platform that worked with the client’s real codebase, documentation, data, and internal tools. It had to be more than a generic chatbot. The system needed specialized AI agents, clear responsibilities, human oversight, and traceable results.
Project duration:
6 months in active phase
Project team:
Solution Architect, AI/ML Engineers, Backend Engineers, Data Engineers, DevOps Engineers, Business Analyst, QA Engineers
Project labor costs:
18 man-months in active phase
Technology stack:
Claude, Gemini, Python, FastAPI, PostgreSQL, ClickHouse, Qdrant, Jira API, CI/CD, MCP, RAG, Microsoft Teams, cloud storage integrations, REST API
The Solution
XIM developed a multi-agent AI system with several connected automation tracks.
For development teams, we built an AI pipeline launched directly from Jira. One agent analyzes the task and creates a grounded plan. It uses project documentation, domain risks, forbidden files, and exact code references. Another agent implements small, focused code changes. A review agent checks the result and decides whether it should be accepted, reworked, or escalated to a human engineer.
After each commit, documentation is updated step by step. This helps the system stay aligned with the codebase and improve over time.
For QA teams, we added agents that inspect the codebase, find areas with weak test coverage, and generate automated tests. These tests are based on real product logic and known risks, not generic templates. Test results are included in the same accept, rework, or escalate flow.
For business teams, we created an AI analytics assistant connected to ClickHouse. It uses a semantic layer with metric definitions, formulas, table relationships, and common analysis patterns. Employees can ask questions in plain language through a chat interface and receive answers, SQL-backed results, or data tables.
For marketing teams, we built AI-powered video creative analysis. The system monitors cloud storage for new ad videos, analyzes them with multimodal AI, and saves structured descriptions. These include tags, summaries, detected text, calls to action, visual attributes, and embeddings for search and analytics.
For IT teams, we created an assistant inside Microsoft Teams. It handles routine administrative requests from Teams, APIs, webhooks, and monitoring systems. The assistant analyzes each request with RAG, routes it to the right agent, and keeps every step traceable.
The Outcome
The AI agentic platform helped the client automate repetitive work across development, QA, analytics, marketing, and IT operations.
Engineering teams could move from Jira tasks to pull requests faster. QA teams gained better test coverage and more reliable validation. Business users received answers to data questions without waiting for manual analysis. Marketing teams got a searchable library of ad creatives. IT teams reduced the load from routine requests and improved response speed.
The result is a practical AI agentic platform that connects tasks, code, tests, documentation, analytics, and operations into one self-improving workflow.