> selma kocabıyık — AI engineer

I build AI systems
that actually work.

AI Engineer based in the Netherlands. BSc AI from VU Amsterdam. RAG pipelines, multi-agent systems, production LLM apps. Currently building vibeapp.

> see my work> watch @selma.builds
scroll ↓
RAG◇LLMs◇Multi-Agent◇MCP◇Claude◇Next.js◇React Native◇pgvector◇Qdrant◇Embeddings◇Groq◇n8n◇FastAPI◇Vercel◇Prompt Engineering◇Semantic Search◇Fine-tuning◇Vector DBs◇RAG◇LLMs◇Multi-Agent◇MCP◇Claude◇Next.js◇React Native◇pgvector◇Qdrant◇Embeddings◇Groq◇n8n◇FastAPI◇Vercel◇Prompt Engineering◇Semantic Search◇Fine-tuning◇Vector DBs◇
> flagship · open source

Loop
Engineering

Stop writing prompts. Start building agent loops.

A Claude Code skill for self-running agent loops — the system works until a real, testable goal is met. A mechanical verification gate (a Stop hook running actual tests, zero tokens) means the AI can never sign off its own half-built work. The maker can't be the judge.

“A reliable loop is worth a thousand perfect prompts.”
Claude CodeStop HookClosed LoopVerification GateAgent OrchestrationMIT
view on github →read the write-up
closed_loop · stop_hook
> production RAG · floorplanner

RAG Pipeline — Design Dataset Search

A production retrieval pipeline over millions of rooms, floors and product placements. I built The Chunker and a hierarchical chunking strategy (project → floor → room → items) so the LLM reasons over the data instead of just searching it.

LangChainQdrantEmbeddingsk-NN RetrievalSemantic SearchHierarchical Chunking
view project →
embedding_space · k-NN
> the knowledge engine · obsidian

The persistent vault behind every video and thread — an LLM agent reads new sources, links them into a living knowledge graph, and feeds the content pipeline. Move your cursor through the nodes.

knowledge_graph.canvas
live · interactive
> product · stackfit

StackFit

in development

A tech-stack matchmaker for developers — answer a short quiz about your goals and constraints, and StackFit recommends a curated stack and learning path for your archetype (from “vibe coder” to full-stack engineer). An interactive catalog maps tools, ecosystems and roles. Built with a dark, terminal-inspired design system.

Next.js 16React 19TypeScriptTailwind 4shadcn/uiZodi18n (TR/EN)
view on github →
> stack
AI / LLM
Claude (Opus/Sonnet/Haiku)OpenAI GPTMistralGoogle GeminiGroqOllamaLangChainLangGraphRAG PipelinesVector SearchEmbeddingssentence-transformersPrompt EngineeringSSE StreamingStructured OutputsFunction Calling
Agents / Claude Code
Claude CodeMulti-Agent SystemsSubagentsAgent OrchestrationSkillsStop HooksLoop EngineeringMCPMCP ServersContext7Playwright MCPOpenClawComputer Usen8n
ML / Data
PyTorchTensorFlowscikit-learnHugging FaceTransformersNumPypandasspaCyNLTKMatplotlibJupyterFine-tuningStatistical Modeling
Web / Mobile
Next.js 15React 19React NativeExpoTypeScriptTailwind CSSshadcn/uiFastAPINextAuth v5ZodFramer MotionGSAP
Data / Infra
PostgreSQLpgvectorQdrantSupabaseDrizzle ORMSQLiteRedisVercelDockerGitGitHub ActionsREST / SSE APIs
Automation / Content
n8nWeb ScrapingBeautifulSoupObsidianHiggsfieldGroq WhisperMarkdownETL PipelinesData Cleaning
0K+
youtube views
0
videos shipped
0
projects featured
0
languages
> building in public · @selma.buildsall videos →
▶breakout · 53K views

Claude’u 4’e Böldüm: Multi-Agent Sistemi Kurdum

4-agent system from scratch — planner, ui-agent, builder, reviewer

▶28K views
Claude Artık Unutmuyor: Hafıza Sistemi
persistent memory · obsidian
▶11K views
MCP’yi Bu Şekilde Kullanıyorum
model context protocol
▶5K views
Bilgisayarımı Claude’a Verdim
computer & browser use
▶4.9K views
İki AI Ajanını Konuşturdum
openclaw + claude code
▶3.4K views
Claude Code’a YouTube Kanalımı Yönettirdim
/youtube inside claude code
▶1.9K views
Claude Code Token’larımı 4’e Böldüm
token efficiency
let's build something
> get in touch