OnurErgüden

What I do

  1. RAG & LLM apps

    Retrieval pipelines that ground language models in real documents, with evaluation that shows where every answer comes from.

    • Python
    • Llama
    • FAISS
    • Hugging Face
    See Course Intelligence

  2. Tool-calling agents

    Agents that call real tools through LangGraph and MCP: guarded loops, schemas on every write and a person in the loop where it matters.

    • LangGraph
    • MCP
    • LangChain
    • Gemini
    See GymRap AI Coach

  3. ML models & evaluation

    Models trained on messy public data, compared honestly and explained with the metrics that matter for the decision.

    • Python
    • scikit-learn
    • XGBoost
    • pandas
    See HealthFactor-AI

  4. Full-stack products

    Web apps from database to interface: Spring Boot services, typed React front ends and real-time updates.

    • Spring Boot
    • React
    • TypeScript
    • PostgreSQL
    See TaskFoo

  5. Mobile apps

    Cross-platform Flutter apps on Firebase, with AI features built into everyday flows.

    • Flutter
    • Dart
    • Firebase
    • Gemini
    See Kuyumcum

Experience

  1. Future Is Now

    AI engineer

    • Full-stack, client-facing AI applications served with FastAPI.
    • Agentic workflows with LangChain and LangGraph over the OpenAI, Anthropic and Gemini APIs.
    • Qdrant-based retrieval architectures for RAG pipelines.
    • Python
    • LangChain
    • LangGraph
    • Qdrant
    • OpenAI API
    • Anthropic API
    • Gemini
    • FastAPI

  2. Future Is Now

    AI engineering intern

    • AI prototypes.
    • Retrieval-based application features.

  3. VBT Software

    Software engineering intern

    • TaskFoo’s project, epic and task workflows.
    • Role-based access control with PostgreSQL persistence.
    • Docker packaging.
    • Spring Boot
    • React
    • TypeScript
    • PostgreSQL
    • Docker
    See TaskFoo

  4. BMC Otomotiv

    Software engineering intern

    • SAP ABAP applications.
    • Warehouse inventory workflows.
    • SAP ABAP
my_tech_stack.exe

Every ball is a tool from my own projects. Hover to name it, drag to move it.

Every tool here comes from my own projects; the list says where.

My tech stack

All technologies (48)

Languages

AI & ML

Web & mobile

Data & tools

Bliss © Microsoft · photo by Charles O’Rear

About me

I’m Onur, an AI engineer at Future Is Now. I build LLM applications end to end: retrieval, tool-calling agents with LangGraph and MCP, and the product around them, mostly in Python, TypeScript and Java.

I care about the decisions behind a system: how it handles uncertainty, how it is evaluated and how people actually use it.

Away from the keyboard, I’m usually on a basketball or tennis court.

Degree
BSc Software Engineeringİzmir University of Economics · June 2026 · GPA 3.30 / 4.00
Academy
Google AI & Technology AcademyDeep Learning · 2026
Publication
IJEAAccepted · publication pending
Contact meRequest my CVMy research

Projects

All projects

  1. Kuyumcum

    Product engineering · applied AI

    Case studyClosed source

    A mobile ecosystem connecting precious-metal holdings, portfolio insight and local merchants. I co-developed the product and engineered its AI features.

    Kuyumcum jeweler discovery mapKuyumcum AI report list

    Built with

    • Flutter
    • Dart
    • Firebase
    • Python
    • YOLO11n
    • TensorFlow Lite
    • Gemini
    Read the case study

  2. HealthFactor-AI

    Machine learning · research

    Case studySource code

    Two layers of water-safety intelligence: detecting contamination now and forecasting how safety may change next.

    Daily and monthly mean Health Factor of İzmir water samples, August 2025 to August 2026
    Correlation matrix of water-quality parameters and the Health Factor
    26,469

    water-quality measurements from 76 sampling points, August 2025 – July 2026

  3. Course Intelligence RAG

    Retrieval · LLM systems

    Case studySource code

    A departmental knowledge assistant combining local language models, semantic retrieval and deliberately constrained answers.

    Course documentsHierarchical chunksSBERT embeddingsFAISS searchFilters + contextLlama 3.1 (8B)Grounded answerRetrieval →← Generation
    Conceptual diagram

    Built with

    • Llama 3.1
    • SBERT
    • FAISS
    • PyQt6
    Read the case study

  4. GymRap AI Coach

    MCP · agent tooling

    Case studySource code

    A personal strength coach on Health Connect, Cloudflare and an OAuth-protected MCP server. Tested, deterministic analysis runs on the server; ChatGPT reads it through MCP and writes the coaching.

    Daily coach report email, rendered from synthetic dataPost-workout report email, rendered from synthetic dataHeart-rate zones and weekly sets per muscle group, rendered from synthetic data

    Built with

    • MCP
    • TypeScript
    • Cloudflare Workers
    • D1
    • OAuth 2.1
    • Kotlin
    Read the case study

Research

My interests sit at the intersection of applied machine learning, data science and reliable AI systems. I want to understand how models behave beyond a single evaluation score.

Accepted · publication pending

Two-layered Artificial Intelligence System to Assess and Forecast the Safety Level of Drinking Water Resources

International Journal of Engineering Approaches (IJEA)

O. Ergüden, B. Ceylani, A. Şengül, S. Yılmaz, E. Yılmaz, M. Y. Kalkan, D. E. Fawzy

A two-layer approach to current contamination detection and future water-safety trends.

Independent study · not submitted for peer review

Clustering and forecasting public transport ridership in İzmir

Three-person CE 477 course project

25,775 daily İzmirim Kart ridership rows by operator and fare type, January 2021 – April 2025. DBSCAN clustering, seasonal features and demand forecasting with XGBoost and Random Forest.

Questions I’m exploring

  1. How should retrieval systems behave when the available evidence cannot support an answer?
  2. How well do predictive models generalize across time and changing data sources?
  3. How can evaluation separate model quality from data leakage and dataset-specific shortcuts?
HealthFactor-AI case studyResearch details

GitHub activity

Loading GitHub activity… See my profile on GitHub

Certificates

  • Training completion certificate from the Google AI & Technology Academy.

  • Professional certificate · six courses covering traditional and agile project management.

  • Training completion certificate from the Google AI & Technology Academy.

  • Record of achievement · 80%. Model compression and low-bit quantization for efficient AI.

  • Participation certificate. CNNs, vision transformers, segmentation and detection with PyTorch.

  • Record of achievement · 78.3%. Software energy consumption and LLM performance measurement.

Deep learning training completion certificate awarded to Onur Ergüden by the Google AI & Technology Academy (Yapay Zeka ve Teknoloji Akademisi) on May 22, 2026.

Deep learning

Google AI & Technology Academy ·

Training completion certificate from the Google AI & Technology Academy.

Open full-size image

Good work starts with a conversation.

Let’s talk

onurerguden5@gmail.com