Rabia Yoruk

Rabia Yoruk · Senior Software Engineer

I build data and LLM systems that run in production, and I study how they fail.

Senior Software Engineer with 7+ years in Python, backend, data pipelines, ML and LLM applications (RAG, agents). From January 2027 I am an M.S. student in Software and Security Engineering at Texas Tech University in Dallas.

Türkiye and the US. Dallas, TX from January 2027.

Portrait of Rabia Yoruk
00

How Scrapus works

Public listings come in on the left. Records that match banned medicines leave on the right, into TEB's systems. Click the LLM stage to test it.

Scrapus pipeline: four kinds of public sources feed the crawlers, records are normalized, an LLM with retrieval classifies them, flagged records go to TEB through a REST API and cleared ones go to an archive. Marketplace A Marketplace B Classified ads Social shops Crawlersscheduled runs Normalizeclean, dedupe LLM classifyRAG · open playground REST APIflagged, to TEB Archivecleared Scrapus pipeline: four kinds of public sources feed the crawlers, records are normalized, an LLM with retrieval classifies them, flagged records go to TEB through a REST API and cleared ones go to an archive. Market A Market B Classifieds Social Crawlersscheduled runs Normalizeclean, dedupe LLM classifyRAG · open playground REST APIflagged, to TEB Archivecleared

Illustrative. Listings are synthetic. The volume is real: about 60,000 records a day.

Simulated 08:00 0 records

One second on screen is one minute of traffic. Each dot is 10 records. The flag rate here is made up.

Latest delivered records

    Playground: rules vs LLM

    The classify stage, up close. A keyword rule list is the baseline. The LLM reads the listing and the retrieved reference entries. Pick a listing.

    Listing

    Slimming capsules, 60 pcs, imported, fast shipping

    Marketplace A · 450 TRY

    Keyword rules

    Flag

    Matched: capsules, slimming

    LLM with retrieval

    Flag confidence 0.94

    Weight loss product sold with health claims and no registration. Products like this often hide drug ingredients.

    reference: unlicensed slimming products

    This page is static. It does not call a model, so your own text only runs the keyword rules.

    Turn on JavaScript to see the other seven listings and the moving pipeline.

    01

    Work

    1. 60,000 records a day on average, collection to API Scrapus Web monitoring for the Turkish Pharmacists' Association (TEB). It finds banned medicine listings on online marketplaces and sends them to TEB's systems. Co-founder · Feb 2025 to now
    2. 82.5% forecast accuracy across about 10 SME clients Kolay.AI demand forecasting Forecasting for small and medium businesses. Models predicted demand swings and disruptions, and the clients used them to plan inventory. Data Scientist and ML Engineer · 2022 to 2023
    3. 1,000 individual investors over four years, approximately XchangeWiser A stock market product for individual investors. Forecasting and simulation showed each user the projected gains and losses of a decision. Data Scientist and ML Engineer · 2022 to 2023
    4. 4 months from zero to first trial users AI cold outreach platform An AI cold outreach platform for sales teams at a US startup. It researches each prospect and drafts a personal email. Senior Software Engineer, AI platform · Mar 2025 to Jun 2025
    5. 20+ client projects delivered by the team I led AutoML platform for a telecom company An AutoML platform for one of Türkiye's largest telecommunications companies. Its data flow module works like KNIME: each data step is a separate node on a canvas. I led the build as tech lead. Tech Lead · 2023 to 2025
    02

    Research

    1. a.

      Reliable LLM systems

      When does an LLM classifier stay correct after the data changes? I measure drift, hallucination and failure modes in systems that run every day.

    2. b.

      Cost and safety of ML in production

      Accuracy is one number. Cost per record, latency and the price of a wrong decision are others. I look for designs that keep all of them in check.

    3. c.

      Web data at scale

      Agents and crawlers that collect public web data, and ways to evaluate them honestly: coverage, freshness and error rates.

    In preparation

    • Cost, accuracy and drift in LLM-based classification of web records.
    • An evaluation framework for agent-based web data collection.
    • Detecting hallucinations in production RAG decisions.
    03

    Papers

    1. [1]

      A review of AI-based business lead generation: Scrapus as a case study

      A. Kaplan, S. E. Seker, R. Yoruk

      Frontiers in Artificial Intelligence, 8, 1606431, 2025.

      Third author doi:10.3389/frai.2025.1606431 BibTeX

      Open access, CC BY. Reports about 90% precision and recall for lead qualification, and extraction F1 rising from about 0.77 to 0.92.

    2. [2]

      An AI-based personalised job recommendation and application assistant agent for enhanced employment matching: a Scrapus use case

      R. Yörük

      Journal of Data Analytics and Artificial Intelligence Applications, 1(2), 172-189, 2025.

      Sole author doi:10.26650/d3ai.1715642 BibTeX

    3. [3]

      Enhancing SME operations with machine learning and business intelligence: a case study of Kolay.ai

      R. Yörük

      Journal of Data Analytics and Artificial Intelligence Applications, 1(1), 61-83, 2025.

      Sole author BibTeX

    4. [4]

      Customer lifetime value (Müşteri Yaşamboyu Değeri)

      R. Yörük, E. Eşmekaya

      YBS Ansiklopedi, 5(2), 26-40, 2018.

      First author BibTeX

    Books, in Turkish

    • İnternet Nasıl Çalışır? (How the Internet Works). S. E. Seker, R. Yoruk. Bilkav, 2018.
    • Başlangıç Düzeyinde C++ Öğrenimine Giriş (Introduction to C++ for Beginners). S. E. Seker, R. Yoruk. Bilkav, 2018.

    All as BibTeX · Google Scholar · ORCID

    04

    Experience

    1. 2025-2026

      Senior Software Engineer, PeK Automotive (Peker Holding) · Slovenia, remote

      Backend services and workflows for a manufacturing ERP. Python, Django, Angular.

    2. 2025-now

      Co-founder, Scrapus · Tallinn, part-time

      Web monitoring product. About 60,000 records a day for TEB, from collection to API.

    3. 2025

      Senior Software Engineer, AI startup · US, remote

      Built an AI cold outreach platform for sales teams, from zero to first trial users.

    4. 2023-2025

      Tech Lead, OptiWisdom · Istanbul

      Led about five parallel client projects, 20+ delivered. AutoML platform for a large telecommunications company.

    5. 2022-2023

      Data Scientist and ML Engineer, OptiWisdom · Remote

      Kolay.AI demand forecasting and the XchangeWiser market simulation.

    6. 2020-2022

      Software Engineer, Backend and Data, OptiWisdom · Remote

      REST APIs, Celery and Redis jobs, PostgreSQL and MySQL tuning.

    7. 2019

      Software Developer, Epigra · Istanbul

      Web apps and CMS projects in PHP and Laravel.

    8. 2018-2019

      Python Developer, Bilkav · Istanbul

      Python tools. Co-wrote two programming books.

    Education

    1. 2027-2029

      M.S. Software and Security Engineering

      Texas Tech University, Dallas, TX. Starts January 2027.

    2. 2024-2025

      Graduate coursework in Data Science

      Galatasaray University, Istanbul. Courses, not a degree.

    3. 2017-2022

      B.S. Software Engineering

      Istanbul Sabahattin Zaim University, Istanbul

    05

    Mentoring and service

    1. 2026-now

      Fellow mentor, ADPList

      Career direction, interview practice and portfolio projects for students and early career developers.

    2. 2025-2026

      Mentor, Türkiye'nin Mühendis Kızları (Engineer Girls of Türkiye)

      Monthly one-to-one sessions with women engineering students and new engineers.

    3. 2023-2025

      Event director, Keep It Up Network, JCI Antalya

      Ran a networking event series for local business owners and investors.

    4. 2019-2020

      Vice chair, IEEE IZU Student Branch

      Company visits, Arduino workshops and technical competitions.

    06

    Contact

    Questions about my work or papers, research collaboration, or a mentoring session: write to me.

    rabiayrk@gmail.com