AI • Computer Vision • Embedded

Edge AI & Computer Vision — real-time agents and local LLM tools.

From ESP32/OpenMV prototypes to offline RAG apps — practical, reproducible, open-source.

  • JumpNet: frame-to-action agent + real-time GUI
  • Edge prototyping on ESP32 / OpenMV
  • VQ-VAE @128×128→ sharper than VAE, up to 60× smaller
JumpNet

Real‑time visual decision system for a one‑button platformer

Image‑to‑Image

Lightweight image‑to‑image pipeline

About

I’m a developer focused on Edge AI, computer vision, and embedded systems.

I’ve been coding since I was 12. Today I study Computer Engineering at RWTH Aachen University, exploring how software and hardware come together to solve real‑world problems.

I build lightweight AI pipelines and deploy them on microcontrollers and SoCs, with a focus on measurable latency and small memory. I also co‑founded Qgen, an EdTech startup making learning more interactive and impactful.

  • RWTH Aachen University
  • Coding since 12
Open to Werkstudent / Internship / Part‑time in AI, Computer Vision, or Embedded Systems.
Portrait of Ertuğrul Mutlu

Projects

Organized by category. Short descriptions, tech badges, and quick links only.

Machine Vision

JumpNet — Vision-Based AI Agent

Real-time visual decision system for a one-button platformer (~18 ms, ~87% accuracy).

PyTorch Behavior Cloning OpenCV Latency: 18 ms

Face Landmarks Detection (OpenCV DNN + Facemark)

Real-time 68-point facial landmark tracker with EMA smoothing and CSV logging.

Python OpenCV (DNN) Facemark LBF EMA CSV

Edge AI & Embedded

Edge AI Sound Classifier on Raspberry Pi Pico

End-to-end TinyML pipeline on RP2040 that classifies short audio into baby cry, doorbell, smoke alarm, and background. 33-dim lightweight features (Goertzel bands + spectral stats), multinomial Logistic Regression (~87% accuracy), and C++ firmware with real-time z-score, softmax, and hysteresis FSM.

Python scikit-learn C++ USB-CDC TinyML

Pico Edge Trend Alarm (RP2040)

Edge AI pipeline for detecting workload-induced heating trends on the Pico: data logging firmware + Python logger, rolling OLS slope features, Logistic Regression, and C++ inference with oversampled ADC, slope buffering, and hysteresis/hold FSM.

Python Logistic Regression C++ Pico SDK TinyML

Generative Models

Compressing Faces — VAE vs VQ-VAE

Identity retention vs compression — continuous latents vs codebooks.

PyTorch VAE VQ-VAE

Conditional GAN — Sketch Generation

Class-conditioned B/W sketch synthesis; projection discriminator experiments.

PyTorch cGAN Computer Vision

Tools & Apps (Data)

Advanced Dataset Viewer & Collector

Record screen+keypress, browse entries, and analyze behavior-cloning datasets.

Python Streamlit OpenCV

PDF Summary Chatbot (Local)

Offline PDF summarizer with page-range selection and export.

Python Streamlit Local LLM

Real-Time Image Color Palette Extractor

Streamlit app extracting dominant colors from images in real-time using k-means, CIE LAB, and ΔE2000.

Python OpenCV k-means CIE LAB ΔE2000

Applied ML & RL

Applied Linear Regression (x5)

Five scikit-learn regressors on real datasets (salary, cars, wine, insurance, trips).

Python scikit-learn Pandas

Machine Learning Algorithms Comparison

Decision Tree, SVM, and KNN evaluated with Accuracy, Precision, Recall, F1.

Python scikit-learn Metrics

Sentence Prediction with Q-Learning

Character-level RL to reach target sentences; simple RL/NLP crossover.

Python Q-Learning NLP

Wavelets & Clustering Parity

Odd/even classification with wavelet features and clustering — a playful study.

PyWavelets Clustering NumPy

Competitive Programming / Algorithms

Daily LeetCode Progress

Ongoing series of algorithm & data structure challenges — daily solutions + 5-day recaps.

Python C++ Algorithms Problem Solving

Systems & Embedded

Mini SCADA Console (Raspberry Pi Pico, C++)

USB-CDC command console simulating an industrial field device: telemetry (internal temp via ADC4), logging, and LED control. PuTTY @115200 for interaction.

C++ Pico SDK USB-CDC ADC CMake/Ninja

Simple Shell in C

Minimal CLI to explore processes, files, and basic system calls.

C POSIX Processes

Skills Matrix

AI / ML

  • PyTorch, torchvision, torchmetrics · NumPy, pandas, scikit-learn
  • Supervised learning (classification & regression), behavior cloning
  • Metrics & logging: Accuracy / F1 / MAE, TensorBoard, matplotlib
  • Data pipelines: loaders, balancing/negatives, augmentations

Generative Models

  • VAE & VQ-VAE (vector-quantized codebooks, commitment loss)
  • Conditional GAN (projection discriminator experiments)
  • Losses: MSE · KL · BCE / adversarial
  • Checkpointing, sample logging, reproducible runs

Edge AI & TinyML

  • Audio features: Goertzel band energies, spectral stats (33-dim)
  • Classical ML on-device: multinomial / binary Logistic Regression
  • Real-time inference budgets (≤ ~20 ms), z-score, softmax
  • Hysteresis & hold FSM; rolling OLS slope features

Embedded (RP2040 / Pico)

  • C/C++ · Pico SDK · CMake/Ninja · arm-none-eabi-gcc
  • USB-CDC stdio, serial I/O (PuTTY @115200)
  • ADC4 (internal temperature), oversampling & buffering
  • SCADA-style command/response console design

Systems Programming

  • POSIX syscalls: fork, exec, wait; basic CLI parsing
  • Files/process management in C; minimal shell architecture

DevOps & Project Quality

  • Git/GitHub workflows, structured READMEs, issue-driven iteration
  • Reproducible environments & clear repo organization

Competitive Programming

  • Data Structures & Algorithms (Python, C++)
  • Problem-solving, coding patterns, daily practice

References & Feedback

Barış Ünver
Founder (AI companies) · Direct manager
Recommendation · Sep 1, 2022 · TURK AI
I worked closely and advised Ertuğrul. During his volunteer experience in our company, TURK AI, I had the opportunity of knowing him. He is a gifted person but beside his skills he is eager to learn and discover. I was surprised how he could adapt and start to create and develop hard tasks. He gained good knowledge and hands on experience on edge computing devices like OpenMV and ESP32 both hardware and software sides. I trust him that he can bring positive impact and joy to his future colleagues. I strongly recommend him!

Contact

Based in Aachen, Germany. Open to remote/hybrid.

Email LinkedIn GitHub CV (PDF)