Skip to content
View 4F71's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report 4F71

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
4F71/README.md

Hi there, I'm Onur TİLKİ

Typing SVG

LinkedIn GitHub Kaggle Email

Profile Views


About Me

class AIMLDeveloper:
    def __init__(self):
        self.name = "Onur TİLKİ"
        self.role = "AI / ML Developer"
        self.location = "Ankara, Turkey"

        self.core_expertise = [
            "End-to-end ML pipelines (data → model → deployment)",
            "RAG-based systems and LLM integrations",
            "Production-oriented development mindset"
        ]

    def current_focus(self):
        return "Designing practical AI systems with real-world constraints"

Tech Stack

Programming & Data Science

Python Pandas NumPy SciPy

Machine Learning & Deep Learning

PyTorch TensorFlow Scikit-learn Keras Hugging Face

Generative AI & LLM Stack

LangChain ChromaDB Gemini LoRA RAG

Databases

SQL MySQL PostgreSQL

Visualization

Matplotlib Seaborn Plotly

Deployment & Tools

FastAPI Streamlit Docker Git Jupyter VS Code

Optimization

CUDA GPU Optimization 4-bit Quantization


Notable Projects

🛒 Instacart Market Basket Analysis

Binary classification for customer reorder prediction

LightGBMDockerStreamlitFeature EngineeringClass Imbalance Handling

  • F1 Score: 0.78 | Recall: 91% | Precision: 69% on 3M+ samples
  • 15+ engineered features (user history, product popularity, temporal patterns)
  • Complete ML pipeline: EDA → Baseline → Feature Engineering → Production
  • Dockerized deployment on Hugging Face Spaces with interactive demo

Model Card GitHub Live Demo


💬 MentorMate - Production RAG Chatbot

Intelligent FAQ system for bootcamp participants

LangChainChromaDBSentence TransformersMultiQueryRetrieverStreamlit

  • Semantic search across 3,232 Q&A pairs with <3 second response time
  • MultiQueryRetriever + MMR for improved retrieval accuracy (~95%)
  • 40+ keyword mappings with hallucination prevention (<5% error rate)
  • Deployed on Streamlit for 150+ concurrent users

GitHub


📈 HousePrices - Production ML Pipeline

Modular regression system with Kaggle validation

scikit-learnFeature EngineeringRidgeCVAutomated Reporting

  • End-to-end pipeline: data ingestion → feature engineering → training → evaluation
  • Custom features (TotalSF, Age, BathCount) improving RMSE by 12%
  • Automated JSON metrics + feature importance visualization
  • Kaggle Score: 0.13049 RMSLE (0.94 R² local validation)

GitHub


View All Repositories


GitHub Statistics

GitHub Streak GitHub Stats

Contribution Activity


Certifications

Certification Project Outcome Year
Akbank Generative AI Bootcamp Built production RAG chatbot (3.2K Q&As, <3s latency) 2025
Google AI Essentials Applied responsible AI principles in LLM systems 2025
Deep Learning with Keras (BTK) Implemented CNN architectures for classification 2025

Let's Connect!

💼 Open to opportunities in:

Machine Learning EngineerNLP EngineerLLM Engineer
AI/ML ResearcherGenerative AI DeveloperMLOps Engineer


LinkedIn Email Kaggle GitHub

📧 Best way to reach me: mehmetonurt@gmail.com


⭐ If you find my projects interesting, feel free to star them!

Last Updated

Pinned Loading

  1. instacart-next-product-recommendation instacart-next-product-recommendation Public

    Instacart Market Basket datasını kullanarak kullanıcıların bir sonraki muhtemel ürününü tahmin eden uçtan uca bir ML projesi. Proje; EDA, baseline, özellik mühendisliği, model optimizasyonu, değerl…

    Jupyter Notebook 2

  2. MentorMate-SSS MentorMate-SSS Public

    Production RAG chatbot serving 3,232 Q&As with <3s latency and ~95% accuracy using LangChain and ChromaDB

    Jupyter Notebook 2

  3. HousePrices-ML-Pipeline HousePrices-ML-Pipeline Public

    Production-ready ML pipeline for regression tasks with modular architecture (0.94 R², Kaggle validated)

    Jupyter Notebook 1