Engineer & AI Developer · RAG Systems · AI Agents · ML Pipelines · Workflow Automation
Python · LangChain · LangGraph · Ollama · Gemini · Streamlit
- 🔍 I build RAG systems, from hybrid retrieval with local LLMs to cloud-based document chatbots
- 🤖 I develop AI agents and workflow automations using LangGraph, LangChain, Make.com, and n8n
- 🔬 I build end-to-end ML pipelines for classification and predictive modelling
- 🔧 I focus on solutions with measurable business impact, automating workflows, reducing manual work, and making AI practically usable
- 🌍 Mentor at ReDI School of Digital Integration teaching data science to newcomers through real projects
- 🇩🇪 Based in Berlin | 🗣️ English · German · Turkish
| Project | Description | Architecture | Tech |
|---|---|---|---|
| 🐼 Pandas RAG Chatbot | Fully local RAG chatbot over the pandas User Guide — scrapes 23 docs, generates ~1,347 synthetic Q&A pairs, hybrid two-stage retrieval | Web scraping → synthetic Q&A generation → ChromaDB (dual store) → HybridRetriever → Ollama LLM | Python · Ollama llama3.1:8b · ChromaDB · LangChain · Streamlit · sentence-transformers |
| 📄 PDF RAG Chatbot | Upload any PDF and chat with it — FAISS-indexed semantic search with Gemini 2.5 Flash, real-time chunk processing and conversation memory | PDF → chunk (1000t/200 overlap) → Gemini embeddings → FAISS → Gemini LLM | Python · LangChain · FAISS · Gemini 2.5 Flash · Streamlit |
| 🔬 Research Agent | Autonomous AI research assistant — generates structured Markdown reports with source citations and human-in-the-loop clarification | LangGraph state machine → live web search → report generation | Python · LangGraph · Ollama · Flask · CLI |
| 📄 DocuExplorerAI | Select any text in a document and get inline AI explanations — multi-provider AI with Google Search grounding, auth, and cloud persistence | React frontend → Gemini API + HuggingFace fallback → Supabase | React · Gemini · Supabase · TailwindCSS |
| Project | Description | Key Features | Tech |
|---|---|---|---|
| ⚙️ Job Process EngineAI | Batch CV-to-job matching and fit scoring — upload positions, get AI-powered match analysis for all of them at once | Batch profile matching · CV parsing · fit scoring via Gemini | React · Node.js · Gemini |
| 🧠 CareerPromptAI | Enter a career goal, get tailored AI prompts with sources — export as CSV or Markdown | Tailored prompts · source citations · CSV/Markdown export | Gemini · TailwindCSS · JavaScript |
| 🛤️ JobPath | IFS-informed job search platform with Kanban board, emotional check-ins, gamification badges and AI job parsing | Kanban · emotional check-ins · gamification · AI job parsing | React · TypeScript · PostgreSQL · Gemini |
| 🧠 JobGenius | Parse job requirements and get AI career-coaching insights | Job requirement parsing · AI coaching | Streamlit · HuggingFace |
| 📝 Prompt Manager | AI prompt library with CRUD, categories and multi-user auth | CRUD · categories · multi-user auth | Streamlit · Flask · SQLite |
| Project | Description | Key Results | Tech |
|---|---|---|---|
| 💧 Pump It Up | Predict water pump operational status in Tanzania — multi-class classification for social impact | ~82% accuracy (LightGBM) | RF · XGBoost · LightGBM · Streamlit |
| 🏠 House Price Prediction | Kaggle competition — predict residential sale prices from 80+ features | 0.126 RMSLE (Kaggle LB) | LightGBM · GridSearchCV |
| 🚢 Titanic Survival | Binary survival classification with feature engineering (title, family, age×class) | 82.7% accuracy | Random Forest · GridSearchCV |
| 🩺 Diabetes Prediction | Predict diabetes onset with interpretable decision trees | ~75% accuracy | CART · Scikit-learn |
| 🎬 Movie Recommender | Hybrid recommender: user-based + item-based collaborative filtering on MovieLens 20M | Top-10 hybrid recs | Pandas · Pearson CF |
| ⚡ Energy Analysis | Hourly energy consumption analysis with pattern detection and correlation | 4 analysis modules | Pandas · Matplotlib |
RAG & AI Systems
ML & Data Science
Full-Stack & Automation
