Computer Engineering · AI · Robotics

Hello.
I'm Mahmut Esat.

I build intelligent systems around machine learning, retrieval-augmented generation, predictive maintenance, robotics, and industrial automation.

View projects
○ ○ mahmut.profile
Mahmut Esat Kolay

Mahmut Esat Kolay

Computer Engineering Student · AI & ML

Selçuk University · Konya, Türkiye

experience.log

Experience

Microsoft

Microsoft Türkiye

AI Summer Intern

Hybrid · Istanbul, Türkiye

Developed an advanced local-first Retrieval-Augmented Generation system using Microsoft Foundry Local for engineering and industrial knowledge workflows. The work combined hierarchical document parsing, hybrid BM25 and dense retrieval, Reciprocal Rank Fusion, cross-encoder reranking, query rewriting, retrieval grading, context reconstruction, citations, entity-graph exploration, and latency telemetry through a FastAPI application layer. It focused on retrieval quality, explainability, and dependable offline execution for technical knowledge environments.

View internship project
TUSAŞ Engine Industries

TUSAŞ Engine Industries

TEI Aviation Engines School

Selected for the TEI Aviation Engines School, a specialized aerospace-engine training program focused on introducing students to aviation propulsion technologies, engine systems, and industry-oriented engineering practices. The program provided structured exposure to aerospace engine concepts and the technical ecosystem surrounding modern propulsion systems.

FAB

FAB Technologies Inc.

Intern – Computer Engineering & Product Management

Remote · Delaware, NY

Completed an AI & Startup Strategy Fellowship, applying Founder and Project Mindset principles to develop ideas from 0 to 1. Built UiPath workflow automations and used Google AI tools to translate business needs into generative-AI-driven solution concepts.

OSR

OSR Robotics

Intern – Computer Engineering & Robotics

On-site · Konya, Türkiye

Designed and implemented ML-driven fault-detection approaches for industrial robotic systems. Programmed KUKA robots and examined robot motion through kinematic analysis and dynamic modeling, connecting machine-learning methods with practical industrial robotics workflows.

Selçuk University

Selçuk University

Bachelor of Computer Engineering

On-site · Konya, Türkiye

Working on projects in Machine Learning, Deep Learning, Data Science, robotics, and intelligent industrial systems while studying core computer engineering topics including algorithms, data structures, object-oriented programming, and databases.

Komalp

Komalp Energy and Power Factor Correction Systems

Intern – Industrial Automation & Electrical

On-site · Konya, Türkiye

Participated in electrical installation works, assembly and maintenance of power factor correction panels, energy efficiency measurements, and system optimization activities. Took an active role in fault detection and repair, circuit connections, and on-site applications.

Devpak

Devpak Packaging Machines

Intern – Industrial Automation & Control Systems

On-site · Konya, Türkiye

Assisted in PLC programming, control panel design, and machine wiring for industrial automation systems. Contributed to the development of electrical systems for packaging and stretch wrapping machines. Observed and analyzed existing automation systems to identify improvements in operational efficiency.

Mehmet Tuza Pakpen Technical High School

Industrial Automation Technologies

On-site · Konya, Türkiye

Worked on IoT-based projects combining foundational software and electronics skills. Studied microcontroller programming, circuit design, PLC systems, and industrial sensors.

competitions/

Competitions & Research

1st Place - Explore Unilever Factory Experience Case Competition | Unilever @HPC

Designed and presented an end-to-end industrial problem-solving model for factory process optimization during a 1-day immersive case study sprint.

Secured 1st place among participating teams by delivering strategic, feasible, and sustainable operational solutions evaluated by Unilever plant managers and leaders.

Unilever Chain Reaction @ U-House

Participated in the Unilever Chain Reaction supply chain simulation at U-House, executing data-driven optimization strategies and dynamic decision-making under competitive constraints.

TÜBİTAK 2209-A Undergraduate Research

Developed an image processing-based optimization project for the detection and classification of oilseeds using deep learning methods.

Emlak Konut – Smart Vertical Transport & Robotic Construction Competition

Ranked in the top 10 out of 60 teams as a national finalist in the Emlak Konut competition, developing an offline local LLM & RAG-driven AI system for vertical transportation and smart construction diagnostics.

BTK Ankara AI Hackathon @ Ankara Teknopark

As part of a team, designed a functional AI-based prototype application within 30 hours under real-world constraints.

projects/

Selected Projects

local-rag.project
PROJECT 01

Local RAG Assistant

Structure-Aware Hybrid RAG for Technical Document Intelligence

A local-first, offline-first document question-answering system designed for engineering and industrial knowledge environments. It parses Markdown, PDF, Word, Excel, and CSV sources into hierarchical knowledge structures instead of relying solely on flat, fixed-size chunks.

Retrieval combines dense semantic search, BM25, Reciprocal Rank Fusion, cross-encoder reranking, query expansion, retrieval grading, and bounded follow-up retrieval. Parent-section reconstruction, entity co-occurrence exploration, source citations, explainability data, and per-stage latency telemetry are exposed through a FastAPI backend with Microsoft Foundry Local inference.

Foundry Local · FastAPI · RAG · BM25 · Dense Retrieval · RRF · Cross-Encoder

turing-machines.project
PROJECT 02

Turing Machines LAB

An experimentation and simulation environment for exploring Turing machines and foundational computational theory. The project provides a practical way to reason about machine states, transition rules, tape operations, and step-by-step computation.

It is structured as a laboratory for observing how formal transition definitions evolve over a tape, making abstract automata concepts easier to inspect and test.

Turing Machines · State Transitions · Tape Operations · Computational Theory

battery-rul.project
PROJECT 03

Li-Ion Battery RUL Prediction with XAI & Adaptive LSTM

Predicts remaining useful life from the NASA PCoE Li-Ion battery dataset using an LSTM and attention architecture. Attention provides explainability by showing which time steps influence the prediction across 4°C, 24°C, and 43°C operating conditions.

An adaptive padding and windowing strategy supports prediction from Cycle 0 and addresses cold-start behavior for short-life batteries such as B0032.

Python · PyTorch · Deep Learning · LSTM · Attention · Explainable AI

Approximately 84% RMSE reduction was reported for the critical B0032 tested case; this is not presented as a universal result.
kuka-dynamics.project
PROJECT 04

KUKA Dynamics Analysis

A desktop engineering application for analyzing the dynamics of industrial KUKA robots. It brings Newton–Euler and Lagrange formulations into a focused workflow for examining robot motion, torque calculations, and workspace behavior.

Engineering visualizations support the interpretation of dynamic and kinematic results, connecting mathematical models with industrial robotics analysis.

Industrial Robotics · KUKA · Newton–Euler · Lagrange · Torque · Workspace Analysis

inventory-optimization.project
PROJECT 05

AI-Driven Inventory Optimization for FMCG

An end-to-end predictive analytics framework for FMCG raw-material inventory optimization. It combines time-series feature engineering and XGBoost demand forecasting with a dynamic safety-stock algorithm under simulated demand volatility.

The optimization layer evaluates inventory decisions against service-level requirements and total-cost behavior within the project’s simulation and test setup.

Python · Pandas · XGBoost

Reported simulation result: 32.6% reduction in total inventory costs while maintaining a 95% service level.
turbofan-rul.project
PROJECT 06

Remaining Useful Life Prediction of Turbofan Engines Using LSTM Networks

Turbofan Engine RUL Prediction

Developed an LSTM-based Remaining Useful Life prediction system for turbofan engines using the NASA CMAPSS dataset. The model learns temporal degradation patterns from multivariate sensor sequences to estimate remaining operating cycles before failure.

The preprocessing pipeline uses moving-average filtering, variance-threshold feature selection, and sliding windows of 50 time steps. Predictions are exposed through a REST/FastAPI interface and evaluated only against the stated unseen test split.

Python · LSTM · CMAPSS · FastAPI · Time Series

RMSE 13.87 on the stated unseen test set.

certificates/

Certificates & Programs

Ankara_AI_Hackathon_Sertifika.pdf.pdf

Ankara AI Hackathon Sertifika

PDFOpen
Aspire-Certificate.pdf

Aspire Certificate

PDFOpen
My Learning - NVIDIA.pdf

My Learning - NVIDIA

PDFOpen
UnileverChain.pdf

Unilever Chain

PDFOpen
UnileverChallenge.pdf

Unilever Challenge

PDFOpen
UnileverSupplyChain.pdf

Unilever Supply Chain

PDFOpen

skills.json

Technical Skills

Data Engineering & Analytics

Power BI · SQL · ETL · Web Scraping · Pandas

AI & Machine Learning

LLM · RAG · Computer Vision · Deep Learning

Languages

Python · C++ · C# · Java

Robotics & Automation

KUKA Sim Pro · KRL · PLC · ROS · IoT

DevOps & Tools

Git · Docker · UiPath · Linux

documents/

Documents

cv.pdf

Curriculum Vitae

CV · PDFOpen
EsatkolayRAGproject.pdf

FoundryRAG Technical Report

Technical Report · PDFOpen
Battery_RUL_Technical_Documentation.pdf.pdf

Battery RUL Technical Documentation

Technical documentation · PDFOpen
MekReader_Technical_Documentation_v2.pdf.pdf

AI-Powered Academic Paper Asisstant

Technical documentation · PDFOpen
NeuralNetwork_Documentation.pdf.pdf

Neural Network Documentation

Technical documentation · PDFOpen
Remaining Useful Life Prediction of Turbofan Engines Using LSTM Networks with Piecewise Linear Target Labeling MEK.pdf.pdf

Remaining Useful Life Prediction of Turbofan Engines Using LSTM Networks with Piecewise Linear Target Labeling

Research paper · PDFOpen
SUTIS[1].pdf.pdf

SUTIS

Project document · PDFOpen
Veri_Madenciliği_Final-Butunleme_Raporu.pdf.pdf

Iron(FE) atomic spectrum data

Report · PDFOpen

highlights/

Highlights

Selected moments from internships, competitions, engineering work, and technical events.

microsoft-office.jpg
unilever-chain-reaction.jpg
unilever-factory-win.jpg
robotics-internship.jpg
ankara-ai-hackathon.jpg
cubesat-space-selfie.jpeg

contact.mail

Contact

new-message.mail