Power your dreams now!
Răzvan Mareș — Senior AI Engineer & Solution Architect
.NET · RAG · Cloud
I design and build enterprise AI assistants, retrieval systems that cite their sources, and the .NET and Kubernetes platforms they run on.
Production-Grade AI, Grounded RAG & High-Load .NET
Systems Built for Enterprise Scale
Selected enterprise engineering deliveries across ERP AI assistants, partitioned .NET and SQL Server event platforms, European banking credit workflows, and self-hosted Kubernetes infrastructure.
razvanmares.ro
What happens when a software engineer over-engineers their portfolio? 10 languages, 6 themes, Clean Architecture — and a story worth reading.
Enterprise Homelab
50+ containers, a Dell R740XD, and an obsession with understanding every layer from bare metal to Kubernetes.
Enterprise AI Assistant
10,000+ employees, one AI assistant. How I built the tool that changed how an entire enterprise works every day.
Sportsbook Platform
Tens of billions of records, real-time load, zero tolerance for downtime. The backend that powers a global sportsbook.
Online Credit Application
10M+ customers, multiple co-borrowers, one fully digital credit flow — for a bank that had never done it online.
Enterprise AI, .NET & Cloud Delivery
Production software architecture combining grounded Hybrid RAG retrieval, Clean Architecture .NET and SQL Server backends, sovereign LLM deployment, and automated Kubernetes operations.
Enterprise AI & Hybrid RAG
BM25 + pgvector retrieval, cross-encoder reranking, RBAC security trimming, source citations, and private Ollama or EU cloud inference.
High-Load .NET, SQL Server & Distributed Data
Clean Architecture in C# / .NET 10, partitioned SQL Server & PostgreSQL, Apache Kafka event pipelines, and low-latency APIs.
Kubernetes (k3s/AKS) & DevSecOps
Multi-node k3s and Azure clusters, declarative GitOps CI/CD, Traefik edge hardening, and Prometheus/Grafana observability.
Architecture Audits & Modernization
2–4 week grounded AI POC sprints, ERP/Fintech API integrations, and incremental strangler-fig migrations without downtime.
Academic Rigor & Production Scale
Combining continuous enterprise .NET, SQL Server, and cloud integration delivery since 2020 with graduate Master of Science in Artificial Intelligence research and ICI Bucharest safety-critical RAG implementations.
Selected Credentials & Production Systems
Frequently Asked Questions on Enterprise AI, Hybrid RAG & .NET
Direct architectural answers on hallucination-resistant RAG pipelines, high-load .NET and SQL Server backends, private LLM data sovereignty, Kubernetes operations, and B2B consulting.
How do you prevent hallucinations in Enterprise AI assistants and Hybrid RAG systems?
Production Hybrid RAG pipelines combine lexical search (BM25 and full-text indexing) with dense vector embeddings in PostgreSQL pgvector, merged through Reciprocal Rank Fusion and re-scored by a cross-encoder reranker. Document-level RBAC security trimming runs before context assembly so users only retrieve authorized records, while multi-stage input/output guardrails enforce inline source citations and refuse unsupported answers.
How do you architect high-load .NET, SQL Server, and event-driven production systems?
Backend platforms are structured around C# and .NET 10 Clean Architecture with explicit domain boundaries, tuned Entity Framework Core queries, and partitioned SQL Server and PostgreSQL schemas designed for high-concurrency transactional workloads. Write-intensive workflows decouple ingestion from processing via Apache Kafka and RabbitMQ event streams, backed by distributed Redis caching and end-to-end OpenTelemetry tracing.
Can Enterprise AI and RAG run privately on-premise or in sovereign EU cloud environments?
Yes. For regulated teams in finance, banking, legal, telecom, or industrial ERP, open-weight language and embedding models run locally through Ollama or vLLM on dedicated GPU/CPU nodes or inside sovereign European Union cloud tenants. Confidential contracts, database records, and internal procedures stay inside your network boundary in alignment with GDPR, NIS2, and EU AI Act governance, with optional multi-provider failover.
How are Kubernetes (k3s and AKS) clusters, security headers, and CI/CD pipelines delivered?
Workloads deploy declaratively to multi-node k3s and Azure Kubernetes Service (AKS) clusters behind Traefik ingress, TLS termination, and strict Content-Security-Policy headers. Continuous delivery pipelines enforce automated unit, contract, bundle-budget, header-rule, and Playwright E2E checks before rollout, paired with Prometheus, Grafana, and Loki observability plus proactive zero-cookie-before-consent GDPR enforcement.
What happens during an architecture audit, AI proof-of-concept sprint, or legacy modernization?
Engagements typically start with a structured architecture and data-readiness audit that profiles SQL Server or API bottlenecks, document quality, and security constraints. From there, we either execute a focused 2–4 week Hybrid RAG proof-of-concept measured against real domain questions or modernize legacy .NET, React, and ERP integrations incrementally using strangler-fig patterns without disrupting live operations.
What production track record and Artificial Intelligence research back your engineering work?
My software career spans web and backend engineering since 2015 and continuous .NET, SQL Server, Blazor/React, and cloud integration delivery since 2020 across telecom, global ERP, high-load sportsbook, and European banking systems. On the AI track, I pursue an MSc in Artificial Intelligence and built from-scratch Hybrid RAG and classical ML pipelines during collaborative research with ICI Bucharest.
How do B2B consulting engagements and senior engineering contracts work?
You work directly with the architect writing and shipping the code — zero agency layers or junior handoffs — contracted cleanly via Mareș-Medar G Răzvan PFA (RazvanApps) for clients in Romania, the European Union, the UK, and the US. Engagements include mutual NDA protection, full source-code and IP ownership transfer, EU/RO e-Factura invoicing, and start with a focused 15- or 30-minute technical discovery call.
Ready to Create
Something Amazing?I'm always excited to collaborate on innovative projects. Let's build something extraordinary together.