Hi there, I’m

Zeynep

Hi there

I'm Zeynep — an AI Engineer with a passion for building intelligent systems that solve real problems. From large language models and automation agents to full-stack applications, I love turning complex ideas into working products.

Below is a selection of projects across corporate work, apps, and AI/ML research.

Corporate Work

Amazon


Mini Idler Office Game

Mini Idler screenshot

A tiny Three.js idle game built for my team to communicate and feel like working together even when remote. A cozy virtual space where team members can hang out, chat, and stay connected throughout the workday. All in-game 2D assets were generated using Assetify, my own asset generation pipeline.


Influencer Affiliate Marketing

Influencer Affiliate Marketing screenshot

End-to-end automation platform for an influencer affiliate marketing program. Event-driven pipeline of 18 Lambda functions: AI content validation (Amazon Bedrock), product eligibility checks against catalog APIs, programmatic storefront creation, ASIN classification, influencer notifications via in-app messaging and SES, AI-generated WhatsApp messages, and cross-account report generation. Step Functions orchestrate validation, DynamoDB Streams trigger downstream automation, SQS throttles upstream API limits. Deployed across 3 AWS regions with CDK and cross-account IAM roles. Replaced a fully manual process — 30+ concurrent campaigns, 70+ influencers across 6 agencies, thousands of monthly orders, zero manual intervention.


Influencer Affiliate Marketing Dashboard

Influencer Affiliate Marketing Dashboard screenshot

Note: Sensitive data has been blurred for confidentiality.

Internal operations dashboard for tracking influencer affiliate marketing campaigns — real-time campaign status, budget tracking, order analytics, influencer management, and automated payment processing.


Legal Automation Agent

Legal Automation Agent screenshot

AI-powered legal automation system that processes regulatory consumer complaints end-to-end. Takes complaint PDFs, extracts key information via OCR (Textract), queries 4 internal databases — orders, refunds, customer service history, shipment tracking (172 fields total) — via SQL-over-S3 (Athena/Glue), classifies cases against a legal decision tree using RAG-augmented AI reasoning (Bedrock Agent), selects and fills the correct defense petition template from a library of 51 templates, generates evidence documents with data tables as proof attachments, and delivers reports via email and tracking tickets. Reduced per-complaint processing time from 3–6 hours to under 5 minutes. Fully serverless, event-driven, with canary deployment and auto-scaling.


TR Retail Ads Dashboard

TR Retail Ads Dashboard screenshot

Note: None of the data shown in this image is real — it is solely for representation purposes.

Unified analytics dashboard consolidating advertising performance data for the Turkey marketplace, replacing a dual-QuickSight setup. 11 pages covering real-time KPIs, trend analysis (DoD/WoW/MoM/QoQ/YoY), lost business opportunity detection, attrition tracking, DSP plan vs actual, negotiation frameworks, and cross-market benchmarking. Supports 8 granularity levels (Marketplace → ASIN, 161K entities), 12 global filters with bulk input, paginated pivot tables with sortable columns, and CSV/Excel export. Static SPA with lazy-rendered Highcharts, no framework dependency.


SIC Crawler

Developed AI-enhanced web crawling pipelines for e-commerce data extraction across numerous SIC (Standard Industrial Classification) verticals. Built specialized crawlers for selection gap analysis — identifying missing product categories and ASINs against competitor catalogs — and merchant onboarding pipelines that automatically discover, qualify, and enrich seller profiles. Crawlers feed into automated data pipelines for catalog coverage reporting, pricing intelligence, and marketplace expansion prioritization.

Apps


InSync

InSync screenshot 1 InSync screenshot 2

Digitalized and couquette version of the card game "The Mind", I loved playing this game with my friends and especially my loved one, Onur - but the game was not available in Turkey. So I made a digital version of it where you can play side by side. You have to line up the cards from the least to the most between 1 and 100 without talking. If you can, then you are forever in sync.


Eda - Prayer Companion

Eda screenshot 1 Eda screenshot 2

A prayer companion app that helps users track their daily namaz (salah) status with an intuitive interface. Features an integrated AI companion powered by a comprehensive Islamic knowledge base — covering prayer times, Quran references, hadith, fiqh rulings, and daily dhikr. The AI assistant provides contextual guidance on prayer-related questions while respecting scholarly consensus across major madhabs.

AI / ML Projects


BranchMind

BranchMind screenshot

A Codex-native multi-agent engineering workspace where autonomous agents collaborate across the software development lifecycle. PM agents decompose features into tasks, developer agents implement on isolated branches, and team lead agents review pull requests and resolve merge conflicts — all while maintaining full repository context. Designed as an operating system for AI-driven software teams rather than a conversational interface. Currently in private development while finalizing core architecture and workflows, with plans to open-source upon stabilization.


Assetify

A 2D asset generation pipeline for game development. Users provide their own API key to generate game-ready sprites, backgrounds, and UI elements through a customizable pipeline. Handles prompt engineering, style consistency, and batch processing to streamline the asset creation workflow for indie game developers.


GossipAI — Adaptive News Delivery

MIS492 — Graduate Project, Boğaziçi University, Spring 2024

Revolutionizing the news consumption experience through the development of an AI-powered, personalized news platform that caters to user preferences by delivering customized, summarized news feeds.


Advanced Customer Churn Prediction

MIS49Y — Machine Learning, Boğaziçi University, Fall 2023 · Submitted to SIU 2024

This research addresses the role of machine learning techniques in enhancing the prediction and comprehension of customer churn within the context of the increasingly competitive e-commerce sector and rising customer expectations.


LocATE — AHP Restaurant Location Recommendation

MIS463 — Decision Support Systems, Boğaziçi University, Fall 2023

This project utilizes Python, Flask, and React to perform an in-depth analysis of New York City's zip code areas, helping users identify the optimal location to open a restaurant based on the category they choose.


Cirrhosis Disease Survival Prediction Model

MIS49Y — Machine Learning, Boğaziçi University, Fall 2023

This dataset utilizes 17 clinical features for predicting survival state of patients with liver cirrhosis. The survival states include 0 = D (death), 1 = C (survival), 2 = CL (survival due to liver transplantation). Our model predicts if the patient will survive without an operation, survive with a transplant, or lose their lives.


This is just a selection — more projects and details coming soon.

Contact

QWERTYUIOP
ASDFGHJKL
ZXCVBNM

Booting...

Starting...