I ship production-grade AI systems
that actually work.

01

AI Systems Architecture

Multi-agent platforms that survive production.

ReAct agents on LlamaIndex, 10+ LLMs orchestrated through a single interface, full RAG with document parsing, chunking, pgvector. End-to-end streaming. Proper observability via OpenTelemetry. The architecture decisions that separate a cool demo from a system your ops team doesn't hate.

Proof Shipped Alva at Alfa Laval — React UI, .NET 8 orchestrator, Python agent, 10+ LLMs.

02

AI-Driven QA

Tests that write themselves, heal themselves, and ship themselves.

Custom AI agents that auto-generate test plans from UI exploration, create Playwright tests from tickets, self-heal failing tests, and move tickets through the pipeline. Integrated into the dev cycle to catch regressions, OWASP top-10 vulns, and performance bottlenecks — before QA even touches them.

Proof 1000+ tests across 6 Mars Petcare apps. Authenticated test pass rate 20% → 100%.

03

Enterprise DevOps

CI/CD pipelines that do the right thing by default.

Unified pipelines across 7 services with smart change detection and parallel deploys. Prod container registry isolation via image promotion. Bicep IaC with RBAC that actually works. Consolidated coverage reporting across .NET and Python. The boring infrastructure that makes the exciting stuff possible.

Proof System Verification — 300% stability gain, 750% faster test execution. Alva — unified 7-service pipeline.