About BlueRobin
BlueRobin started in 2022 as a simple question: “Why does it take so long to find information in my family’s medical records?”
The Problem
After years of accumulating medical documents—lab results, doctor’s notes, prescriptions, imaging reports—I found myself spending hours searching through PDFs and scanned documents whenever I needed specific information. Traditional file organization wasn’t scaling, and searching by filename was useless for medical content.
The Journey
Phase 1: Cloud-First (2022)
Like many developers, my first instinct was to build on cloud services. I started with:
- Azure Blob Storage for document storage
- Azure Cognitive Search for full-text search
- A simple ASP.NET Core API
- React frontend
It worked, but the monthly costs added up quickly, especially for AI-powered features. For a personal project, $50-100/month felt excessive.
Phase 2: The Homelab Pivot (2023)
I had an old server collecting dust. Why not run everything locally?
This sparked a complete architecture redesign:
- k3s for lightweight Kubernetes
- MinIO replacing Azure Blob Storage
- PostgreSQL with pgvector for embeddings and search
- Flux for GitOps deployments
- Ollama for local LLM inference
The homelab approach forced me to think carefully about resource constraints, which led to better architectural decisions.
Phase 3: AI Integration (2024-2025)
The explosion of accessible LLMs changed everything:
- OCR with Docling for document text extraction
- Local embeddings with sentence-transformers
- RAG pipeline for natural language queries
- Hybrid AI architecture using both local Ollama and cloud APIs
What BlueRobin Does Today
BlueRobin is a personal document management system that:
- Ingests documents - Upload PDFs, images, or scans
- Extracts content - OCR and text extraction with AI cleanup
- Generates embeddings - Vector representations for semantic search
- Extracts entities - Named entity recognition for people, dates, medications, etc.
- Enables search - Both keyword and semantic search with RAG-powered Q&A
The Tech Stack
| Component | Technology | Why |
|---|---|---|
| Frontend | Blazor Server | Real-time UI, C# everywhere |
| API | FastEndpoints | High-performance, clean |
| Database | PostgreSQL + pgvector | Relational + vector in one |
| Messaging | NATS JetStream | Lightweight, powerful |
| Storage | MinIO | S3-compatible, self-hosted |
| AI | Ollama + OpenAI | Local-first, cloud fallback |
| Orchestration | k3s + Flux | GitOps on homelab |
Why This Blog?
Four years of building BlueRobin taught me patterns I wish I’d known earlier:
- How to implement DDD in a real .NET project
- Event-driven architecture with NATS
- GitOps with Flux for homelab deployments
- Practical RAG pipeline implementation
- Balancing local and cloud AI
This blog documents those learnings—with real code from a production system.
About Me
I’m Victor, an Engineering Director who codes on weekends. BlueRobin is my playground for exploring patterns I don’t always get to use at work: DDD, event sourcing experiments, LLM integration, and homelab DevOps.
The best way to learn is to build, and the second best way is to teach. This blog is both.
Have questions about BlueRobin or want to discuss patterns? Reach out on GitHub or LinkedIn.