Engineering Notizen & Open-Source Arbeit

Dokumentation von Engineering-Entscheidungen, Architektur und Erkenntnissen aus dem Aufbau von KI-Systemen in der Produktion.

Blog Articles

Using RAG for SQL Generation — Why Embeddings Beat Prompt Stuffing

8 min read

Jan 28, 2026

How pgvector embeddings improve LLM-to-SQL accuracy by providing schema context instead of dumping entire schemas into prompts.

rag sql generationpgvector embeddings sqltext to sql pgvector

Building an MCP Server with Spring Boot — A Practical Guide

10 min read

Jan 20, 2026

Implementing the Model Context Protocol for AI assistant tool integration using Spring Boot and Spring AI.

mcp server spring bootmodel context protocol tutorialspring ai mcp

Offline-First PWA Patterns — Service Workers, IndexedDB, and Background Sync

7 min read

Jan 15, 2026

Service workers, IndexedDB, and background sync patterns used in MicroItinerary for reliable offline-first travel planning.

pwa offline sync indexeddbservice worker background syncoffline first web app

Beitrags-Feed

rohitraj.tech Redesign

open source

backendscale·2026-01-31

Complete redesign from consultancy landing page to engineer-centric project directory with documentation-first aesthetic.

→ Live at rohitraj.tech

src/appsrc/componentssrc/data

GitHub →

MicroItinerary — AI Travel Planner

ai systems

MicroItinerary·2026-01-27

AI-powered PWA for planning annual travel itineraries. Features intelligent destination suggestions via OpenAI GPT-4, cost estimation in INR, and Splitwise-style expense splitting for group trips.

→ Full-stack PWA with offline support, Google OAuth, Redis caching

backendwebdocker-compose.yml

GitHub →

StellarMIND — Chat-to-SQL with pgvector

ai systems

spring-ai-mcp-server·2026-01-25

Spring Boot MCP server that converts natural language questions into read-only SQL using LLM. Retrieves schema context via pgvector for RAG-based query generation.

→ Natural language → SQL with safe read-only execution

stellarmind-serverstellarmind-client

GitHub →

AWS SQS Integration Patterns

open source

sqs·2021-12-27

Reference implementation for AWS SQS integration with Spring Boot including producer/consumer patterns, retry logic, and dead-letter queue handling.

producerconsumer

GitHub →

Repository Übersicht

backendscale

TypeScriptactive

Engineering work visibility — portfolios show results but not thinking

Full implementation: Next.js 16, React 19, documentation-first design system

src/appsrc/componentssrc/data

MicroItinerary

Javaactive

Travel apps lack intelligent suggestions and group expense management

Full-stack: Spring Boot 3.2.2 + Java 21 backend, React 18 + Vite PWA, OpenAI integration

backendwebtest

spring-ai-mcp-server

Javaactive

Natural language database queries with retrieval-augmented context via pgvector

Full MCP server implementation, pgvector integration, Chain-of-Thought UI

stellarmind-serverstellarmind-clientpostman

my-expense

Javamaintenance

Simple expense tracking without bloated features

Full-stack Java application with REST API

apipersistencereports

sqs

Javaarchived

Reference implementation for AWS SQS with Spring Boot

Producer/consumer patterns, retry logic, DLQ handling

producerconsumerconfig

wealwa-api

Javaarchived

Personal wealth tracking and portfolio management

REST API design, portfolio analytics

apiportfolioanalytics

Alle Repositories auf GitHub anzeigen →

Projekt Notizen

`backendscale (rohitraj.tech)`

frontend

Warum Es Existiert

Engineering work is often invisible. Traditional portfolios show polished results but not the thinking process.

Technische Kernherausforderung

Creating a documentation-first design system that feels like internal engineering docs while remaining visually professional.

Architektur Überblick

Next.js 16 with App Router for static generation where possible
React 19 with server components for optimal performance
Tailwind v4 for utility-first styling
Data-driven content structure in src/data/ for easy updates

Kompromisse & Entscheidungen

Chose vanilla CSS over component libraries for full control, at cost of development speed
Static generation means content updates require redeployment
Minimal dependencies reduce maintenance burden but limit feature velocity

Production — deployed on VercelView Repository →

`MicroItinerary`

ai systems

Warum Es Existiert

Travel apps lack intelligent suggestions based on season/budget and don't handle group expense splitting well.

Technische Kernherausforderung

Integrating OpenAI GPT-4 for destination suggestions and cost estimation while maintaining offline-first PWA capabilities.

Architektur Überblick

React 18 + Vite PWA frontend with IndexedDB for offline support
Spring Boot 3.2.2 + Java 21 backend with Flyway migrations
PostgreSQL 16 + Redis caching layer
Google OAuth 2.0 + JWT authentication
OpenAI GPT-4 API for AI features

Kompromisse & Entscheidungen

PWA over native app — broader reach but limited device APIs
OpenAI API adds latency and cost vs simpler rule-based suggestions
Expense splitting algorithm prioritizes simplicity over Splitwise feature-parity

Development — backend and frontend functional, needs polishView Repository →

`StellarMIND (spring-ai-mcp-server)`

ai systems

Warum Es Existiert

Business users need to query databases without knowing SQL. Existing tools lack context-aware query generation.

Technische Kernherausforderung

Retrieval-augmented SQL generation using pgvector for schema context, with strict read-only query enforcement for safety.

Architektur Überblick

Spring Boot MCP server with Tool interface for query execution
pgvector for storing schema knowledge chunks and embeddings
Spring AI for LLM integration (provider-agnostic)
Chain-of-Thought (CoT) web interface for query debugging
Read-only SQL enforcement (only SELECT, WITH allowed)

Kompromisse & Entscheidungen

Read-only restriction limits use cases but ensures safety
pgvector requires PostgreSQL — not database-agnostic
MCP transport (stdio) limits deployment patterns vs HTTP

Development — core query flow working, needs UI polishView Repository →