Enterprise GenAI platform

Overview

The client is a global industrial company and the one I’ve worked with longest. Over four projects I worked on test automation, DevSecOps and AI engineering.

---

Project 4 · Enterprise GenAI platform · AI Systems Architect

I worked on the architecture and development of the company’s enterprise GenAI platform - a system where 10+ AI models, including OpenAI, Claude, Meta and others, work together to understand company documents and answer questions in real time.

The platform is live in production and available across the organization.

A few things I worked on

• An intelligent agent that selects the right AI model for each task instead of relying on a single model
• Document understanding across PDFs, Word documents, spreadsheets and presentations, grounding responses in actual company knowledge
• Real-time streaming responses, end to end, directly to the user
• Enterprise cloud infrastructure with security, monitoring and automated testing built into the delivery process

We’re now taking this further by moving toward a fully agentic development process - using AI agents to automate the software development lifecycle itself, with an optional human quality gate at the end.

Under the hood

• A ReAct-style reasoning agent on LlamaIndex with FastAPI, orchestrating 10+ LLMs (GPT-4o, Claude Sonnet/Opus, Mistral, Llama, Grok, DeepSeek) through Azure AI Foundry
• Full RAG pipeline: document parsing (PDF, DOCX, PPTX, XLSX, OCR) → chunking → Azure OpenAI embeddings → pgvector retrieval
• .NET 8 orchestrator following Clean Architecture, brokering streaming responses end-to-end (Agent → API → React UI)
• Azure infrastructure: fully defined in Bicep IaC, with per-service RBAC, image-promotion pattern for ACR prod isolation, async Azure Functions for document embedding
• Observability: OpenTelemetry tracing, Application Insights, consolidated coverage across .NET (xUnit) and Python (pytest)
• Unified CI/CD across 7 services: single pipeline file with Detect / Build / Deploy / E2E stages, selective-component builds, parallel fan-out, QA sign-off gates
• Admin Configuration Panel: agents, models, terms, policies CRUD (React + .NET API)

---

Project 3 · Azure DevSecOps · Senior SDET / DevSecOps

I did DevSecOps work across the company’s Azure environment.

• Set up secure CI/CD pipelines with automated tests and vulnerability scans on every deploy
• Configured Azure DevOps agents for Windows, Linux and containerized workloads
• Worked with security specialists to add OWASP and supply-chain checks to the pipelines
• Hardened test automation for regulated environments

---

Project 2 · Reference framework · Senior SDET

I built an automation framework that later projects in the company used as their starting point.

• Developed a C# + Playwright framework for rapid UI and API validation
• Ran it in Azure DevOps CI/CD with parallel execution and pull-request gated builds
• Shortened release cycles by running automated checks on every merge
• Shared the framework with the wider engineering organization as a template

---

Project 1 · Automation rescue · Test Automation Engineer

My first project there was fixing an old, unreliable test automation setup.

• Migrated from legacy tooling to a Selenium-based C# framework
• Redesigned how tests were structured, run and managed
• Set up Azure DevOps pipelines so teams got test results quickly
• Brought stability up ~300%, execution time down 775+% and initial setup time down by another 500%

---

Looking back

Each of the four projects was a bit bigger than the last. I started in test automation and now work on AI systems architecture.