Atalay Aygül

Data Scientist & AI/ML Engineer

LLM Agents · Computer Vision · Time‑Series

Open to remote opportunities

Based in İzmir. At ATDERA I designed and built the company CRM and its autonomous LLM-agent layer, and I run the SEO/GEO programme and every Meta Ads campaign. Open to remote data science and AI/ML engineering roles.

Now
Manager · Operations & Growth lead, ATDERA
Based
İzmir, Türkiye
Scholarship
Google AI & Technology Academy · 2024–25
Works
On-site, hybrid or remote
ask my CV

Hi, I'm ATALAY.AI. I answer questions about Atalay's work, skills and availability, using only the CV on this page.

  • ASKany question about Atalay's work, or just type it
  • PITCHa 30-second pitch for recruiters
  • MAILe-mail Atalay · CV to request the full CV
  • HELPall commands · or tap a suggestion below
Fig. 1 — ATALAY.AI, a guard-railed LLM assistant that answers recruiters from this CV. How it works

Proof

  • 57%

    of atdera.com's Google impressions come from the four-language content cluster I built and optimised

  • ~90%

    less time on weekly reporting after my Python automation

  • 99.4%

    Xception transfer learning · no observed overfitting

    accuracy of the FoliumCare disease-detection model

  • 25K+

    250+ hours of hands-on training · TensorFlow, Vertex AI, Gemini API

    applicants — selected for the Google AI & Technology Academy scholarship

01

Selected work

Four case studies and more · 2024 — 2026

Case 01In production · 2026

ATDERA CRM

ATDERA sales CRM with an autonomous LLM-agent layer

Fig. 2 — Lead path through the CRM, with call outcomes reported back to the ad platform.

The sales and operations CRM I designed and built from scratch — the company's whole lead-to-enrolment pipeline runs on it. One intake pipeline for Meta instant forms, the website and manual entry (deduplication, phone and country normalisation, response-time SLAs); an autonomous LLM-agent layer that qualifies inbound international leads, drafts outreach in the lead's language (email, WhatsApp, call scripts) and routes them into the sales pipeline; call-outcome tracking with an automated follow-up cadence; a CRM→Meta Conversions API loop that reports call outcomes back to the ad platform; instant .docx generation from the opportunity record, a finance/cash-desk bridge, and row-level security with role-based access on Supabase.

  • Next.js
  • TypeScript
  • Supabase / PostgreSQL
  • Vercel
  • Meta Conversions API
  • LLM Agents
Case 02Live · 2026

SEO / GEO

atdera.com — organic lead channel in four languages

Content cluster
53 articles
Languages
ENFRRUAR
Google impressions
57% from the cluster
Google AI impressions
52% from the cluster
Landing pages
30 indexed in 3 days
Rankings
first US #1s
Fig. 3 — The programme at a glance.

The SEO/GEO programme I run for a four-language medical-education site (EN/FR/RU/AR): technical SEO, a 53-article content cluster, structured data, hreflang, Search Console-driven iteration and AI-search (generative engine) optimisation. The cluster now drives 57% of the site's Google impressions and 52% of its Google AI impressions, with the site's first US #1 rankings; 30 new observership landing pages were fully indexed within three days of launch.

  • Next.js
  • JSON-LD
  • Google Search Console
  • GA4
  • SE Ranking
  • Cloudflare
Case 03Live on this site · 2026

ATALAY.AI

Guard-railed LLM recruiter assistant

  1. Input pre-filterobvious injections never reach the model · per-IP rate limits
  2. NVIDIA Nemotronon Cloudflare Workers AI, answering from the CV
  3. Deterministic output lockscanary · prompt-leak · contact-data · negative-claim
  4. LLM judgefail-closed: no verdict, no answer
  5. Answerplain text, from the CV, on topic
Fig. 4 — Request path: an answer is shown only after both checks pass.

The ASK command on this page: a Cloudflare Pages Function running NVIDIA Nemotron on Workers AI, wrapped in an input pre-filter, deterministic output locks (canary, prompt-leak, contact-data and negative-claim checks), a fail-closed LLM judge, per-IP rate limits and a 41-case red-team suite — so it sells the CV, stays on topic and does not fall for prompt injection.

  • LLM Guardrails
  • Cloudflare Workers AI
  • Prompt-Injection Defence
  • LLM-as-Judge
  • JavaScript
Case 041st place · 2025

FoliumCare

AI grapevine leaf disease detection — 99.4% accuracy

Fig. 5 — From photos collected in Manisa vineyards to a mobile app.

Classifies grapevine leaf disease from a single photo. Our team collected and labelled thousands of images in Manisa vineyards with an agricultural engineer, trained an Xception transfer-learning model to 99.4% accuracy with no observed overfitting, and served it through a FastAPI backend to a Flutter app — full pipeline from database to mobile UI. Selected first among all senior projects in the department.

  • Python
  • TensorFlow / Keras
  • Xception
  • OpenCV
  • FastAPI
  • Flutter

More work

Datathon 2025

2nd place · 2025

24-hour time-series challenge on U.S. hospital data

Second place among all teams in a 24-hour datathon organised by the DEU CS-AI community. Three of us worked a time-series dataset of U.S. hospital records — patient risk flags and multi-disease classification. Most of the value came from understanding what the series actually represented before modelling, then heavy feature engineering and model iteration under time pressure.

  • Time Series
  • Multi-label Classification
  • Pandas / NumPy
  • Scikit-learn
  • Feature Engineering

Quant pipeline

In progress · 2026

Deterministic sports-analytics pipeline with an LLM reasoning layer

A modular Python pipeline that turns raw match data into probabilistic recommendations: Selenium/BeautifulSoup scraping, Poisson and Dixon-Coles goal models on rolling xG, expected value and Kelly-criterion staking, and a structured JSON layer consumed by an LLM. The statistical layer stays deterministic and testable; the AI layer only interprets its output.

  • Python
  • SciPy
  • Poisson / Dixon-Coles
  • Kelly Criterion
  • Selenium
  • LLM

Restaurant management system

2024

A payment calculation and financial recording application built for day-to-day use in a restaurant, developed as a three-person team.

  • C#
  • SQL

Library manager

2024

Command-line library management system

A simple command-line library manager written in Python: add, remove and list books.

  • Python
  • CLI
02

Experience

From local IT support to a regional role, then technical and AI owner at ATDERA.

  1. Jul 2026 — Present

    ATDERA

    Jul 2026 — Present

    İzmir, Türkiye · Full-time · On-site

    Manager

    Jul 2026 — Present

    Operations & Growth lead · owner of the CRM, SEO/GEO and Meta Ads

    • Lead operations for a London-positioned medical education and international patient coordination platform: coordinate postgraduate medical training courses across dozens of countries — course directors, partner clinics, contracts and visa logistics — and manage a call-centre team and the lead-to-enrolment pipeline.
    • Designed and built the company CRM from scratch (Next.js, Supabase/PostgreSQL, Vercel) and its autonomous LLM-agent layer: qualifies inbound international leads, drafts multilingual outreach (email, WhatsApp, call scripts) and routes them into the sales pipeline — plus Meta lead-form webhook ingest, call-outcome tracking, document generation from the opportunity record, a finance/cash-desk bridge and row-level-security hardening on the company's Supabase projects.
    • Own SEO/GEO for atdera.com (EN/FR/RU/AR): technical SEO, a 53-article medical-education content cluster in four languages, structured data, hreflang and AI-search (generative engine) optimisation — the cluster now drives 57% of the site's Google impressions and 52% of its Google AI impressions, with the site's first US #1 rankings; 30 new observership landing pages were fully indexed within three days of launch.
    • Run every Meta Ads campaign end to end — audiences, creatives, instant lead forms, Pixel + Conversions API tracking — and built the CRM→Meta feedback loop that reports call outcomes back to the ad platform to improve lead quality.
    • Built and maintain a 345-institution database across 45 countries and multiple surgical specialties, with CRM tracking and KPI dashboards used for partnership targeting.
    • Own the analytics stack end to end: funnel analysis, paid-acquisition performance (Meta Ads), Google Search Console and GA4 reporting, pricing benchmarking and cohort reporting.
  2. Nov 2024 — Jul 20261 yr 9 mo

    Expeditors International

    Nov 2024 — Jul 2026

    İzmir, Türkiye · Full-time · On-site · 1 yr 9 mo

    Global Tech Associate Specialist

    Jan 2026 — Jul 2026

    Promoted from Local I.S. Specialist to a regional role

    • Primary IT point of contact for 14 branches across 9 countries — managed incidents and requests in ServiceNow and routed complex cases to specialist teams.
    • Wrote Excel/VBA and Python automation scripts for business teams beyond the core support scope, and submitted process-improvement scripts to the development team.
    • Supported warehouse operations with custom tooling and data fixes.

    Service Provider Manager

    Oct 2025 — Jul 2026

    Concurrent role

    • Automated weekly performance scorecards for all freight departments (Ground, Air, Ocean) with custom Python — cut manual reporting from 2–3 hours to 10–15 minutes per cycle.
    • Managed service provider relationships and enforced operational performance standards against contractual KPIs.
    • Turned raw operational data into recurring reports used by department managers for provider reviews.

    Local I.S. Specialist

    Nov 2024 — Jan 2026
    • Hands-on IT support for 400+ users: system administration, Active Directory management and hardware/software troubleshooting — average ticket resolution around 10 minutes.
    • Started writing Excel and Python scripts for recurring operational tasks, which grew into the reporting automation work in later roles.
  3. Dec 2024 — Aug 20259 months

    Google AI & Technology Academy

    Dec 2024 — Aug 2025

    Remote · Scholarship programme · 9 months

    Scholarship Recipient — Data Science

    Dec 2024 — Aug 2025

    Selected from 25,000+ applicants nationwide

    • Programme run by Google together with the Turkish Ministry of Industry and Technology, the Presidency's Digital Transformation Office, T3 Foundation and TGV.
    • 250+ hours of hands-on training in TensorFlow, Vertex AI, BigQuery ML and the Gemini API.
    • Built and evaluated predictive models on real-world datasets across finance, healthcare and retail; competed in the graduation bootcamp, datathons and ideathons; additional coursework in English for developers, law, finance and HR.

Earlier roles

  • Data AnalystTurnkey Health Tourism · İstanbul · RemoteJun 2024 — Sep 2024
  • IT SpecialistMedent AcademyAug 2022 — Dec 2023
  • Software Engineer InternRF 304 · KayseriJun 2020 — Aug 2020
  • InternMinistry of Agriculture and Forestry · HatayJun 2019 — Sep 2019
  • InternHatay Metropolitan MunicipalitySep 2016 — Jun 2017
03

About

From support tickets to LLM agents.

I started writing code in a vocational high school web programming track, went on to an associate degree in computer programming at İskenderun Teknik University — graduating first in the department — and transferred to Dokuz Eylül University through the DGS exam to complete a bachelor's in computer science with a focus on machine learning and computer vision.

My senior project, FoliumCare, detects grapevine leaf disease from a single photo. Our team collected and labelled thousands of images in Manisa vineyards alongside an agricultural engineer, trained an Xception transfer-learning model to 99.4% accuracy, and shipped the full stack — FastAPI backend to Flutter app. It was selected first among all senior projects in the department.

In parallel I spent close to two years at Expeditors International, going from local IT support to a regional role covering 14 branches in 9 countries. The scripts I kept writing on the side — Python that cut a 2–3 hour weekly reporting cycle to 10–15 minutes — became my real job description. During the same period I was selected from 25,000+ applicants for the Google AI & Technology Academy scholarship and completed 250+ hours of hands-on training.

Since July 2026 I've been running operations at ATDERA, where I also own growth and the technical side of the business: I designed and built the company CRM (Next.js, Supabase/PostgreSQL) and its autonomous LLM-agent layer that qualifies and reaches out to international leads, I run the SEO/GEO programme that turned a four-language site into an organic lead channel, I manage every Meta Ads campaign, and I maintain a 345-institution database across 45 countries plus the full analytics stack. I'm open to remote data science and AI/ML engineering roles.

Atalay Aygül smiling next to Mia, a small white dog
With Mia.
04

Education

  • 2021 — 2025
    BSc, Computer Science
    Dokuz Eylül University · İzmir

    Focus on machine learning and computer vision. Senior project FoliumCare selected first in the department; 2nd place in Datathon 2025 with the DEU CS-AI community. Coursework: machine learning, deep learning, computer vision, data structures and algorithms, databases, software engineering.

  • 2018 — 2020
    Associate Degree, Computer Programming
    İskenderun Teknik University (İSTE) · GPA 3.29

    Graduated first in the department. Transferred to Dokuz Eylül University through DGS, the national vertical transfer exam. Coursework: programming fundamentals, object-oriented programming, databases, web development.

  • 2015 — 2017
    Web Programming
    Ali Sayar Vocational and Technical Anatolian High School

    Where the coding started.

05

Certifications

Google — Advanced Data Analytics track (Coursera)

CertificateIssuedCredential ID
Google Advanced Data AnalyticsJul 2025U3ZJ9WLAHIOT
Foundations of Machine LearningMay 20258DJFF5S8OK2H
Regression Analysis: Simplify Complex Data RelationshipsMar 2025LOOGG86PDYRV
The Power of StatisticsMar 20256DFN51M46HZ8
Go Beyond the Numbers: Translate Data into InsightsMar 20254EIPZE85DV8D
Get Started with PythonMar 2025W45PSFQNAZ9S
Foundations of Data ScienceFeb 202563Y0XT1DO0IG

Google AI & Technology Academy (Oyun ve Uygulama Akademisi)

CertificateIssuedCredential ID
AI & Technology Academy Bootcamp CertificateAug 202555569141612452
Entrepreneurship Training CertificateAug 202547524452096134
English for Developers (YZTA)Aug 202535425832929674
Datathon CertificateMay 202581081331805071
YZTA 4.0 Ideathon CertificateMar 202516539053490122

Earlier courses (2024 — 2025)

Coderspace Data Science & AI Summer School (Sep 2024) · Global AI Hub: Akbank Python Bootcamp, Introduction to Python, Python for Machine Learning, Introduction to Machine Learning, First Step into AI (2024) · Turkcell Geleceği Yazanlar: Python Programming 101–401, Machine Learning 101, Data Literacy (2024) · LinkedIn Learning: Programming Foundations: Beyond the Fundamentals, Level Up: Python, Data Science Foundations: Fundamentals (2025)

06

Skills

Languages

  • Python
  • SQL
  • JavaScript / TypeScript
  • Java
  • C#
  • C / C++

Machine learning & AI

  • TensorFlow / Keras
  • PyTorch
  • Scikit-learn
  • OpenCV
  • LLM Agents
  • Computer Vision
  • Time Series
  • Deep Learning

Data & analytics

  • Pandas
  • NumPy
  • SciPy
  • Feature Engineering
  • KPI Dashboards
  • Funnel & Cohort Analysis

Cloud & MLOps

  • Vertex AI
  • BigQuery ML
  • Gemini API
  • Google Colab
  • FastAPI
  • REST APIs
  • Next.js
  • Supabase / PostgreSQL
  • Vercel
  • Cloudflare Workers AI

Tools & platforms

  • Git
  • Flutter
  • Selenium
  • BeautifulSoup
  • Excel / VBA
  • ServiceNow
  • Active Directory
  • Meta Ads (Pixel + Conversions API)
  • SEO / GEO
  • Google Search Console
  • GA4
  • CRM Systems

Professional

  • Operations Leadership
  • Growth & Lead Generation
  • Team Management
  • Cross-functional Communication
  • Process Automation
  • Project Coordination
07

Off the clock

When I'm not shipping models.

Chess

Pattern recognition with a clock running — usually a few games a day. chess.com/member/atalayaygul (opens in a new tab)

3D printing

Designing and printing my own parts — the same iterate-measure-fix loop as training a model, just slower and with more plastic.

08

Contact

Let's talk.

I'm open to remote data science and AI/ML engineering roles, and happy to talk about anything involving LLM agents, computer vision, or messy operational data. For my full CV or a phone call, drop me an email.

Email me Request CV

İzmir, Türkiye · On-site, hybrid or remote