AI & Data Developer
NURSENA BAYKIR
Building intelligent systems that actually work in production — agentic workflows, RAG pipelines, real-time stream processing, and predictive models. Currently crafting AI solutions at Brainy.
Building intelligent systems that actually work in production — agentic workflows, RAG pipelines, real-time stream processing, and predictive models. Currently crafting AI solutions at Brainy.
Building AI-powered applications for internal teams and external clients. Developed a live Google Reviews monitoring and auto-response system — an AI pipeline filters incoming reviews, drafts contextual replies, and instantly routes critical complaints (e.g. health incidents) to branch managers via email. Connected company databases, Meta analytics, and third-party data sources into a unified internal chatbot with multi-agent orchestration. Also built automated report-generation agents that compile insights and deliver them on a schedule.
Participated in a competitive AI bootcamp co-run by UNDP and Samsung. Built an anomaly and violence detection system using YOLOv7-based pose estimation, identifying abnormal human behaviors in surveillance footage. Published the dataset on Kaggle.
Designed and implemented an LSTM-based time series model to forecast airport check-in passenger volumes, giving operations teams a data-driven basis for resource planning and staffing decisions.
Worked within the DevOps team reviewing CI/CD pipeline configurations and maintaining Grafana alerting dashboards. Developed a Python-based load testing script with Locust to benchmark system throughput and identify performance bottlenecks under high traffic.
Accepted into TÜBİTAK's university research grants programme in my second year. Led the design of a voice-activated sensor system for visually impaired individuals, integrating audio feedback to assist with spatial navigation and obstacle awareness.
Automatically collects Slack channel messages and generates concise summaries using Claude AI, then posts them back to a designated channel. Configurable timeframe, supports scheduled daily standups and activity digests.
View on GitHub →RAG system that answers questions by combining a local SQLite news database with live internet search. Uses FAISS semantic search (0.7) and BM25 keyword matching (0.3) via an ensemble retriever for optimal recall.
View on GitHub →Stream processing pipeline for real-time analysis of user behavior and product recommendations on an e-commerce platform. Apache Flink consumes Kafka events from a Python producer and processes them in real time.
View on GitHub →Violence and anomalous behavior detection via YOLOv7 pose estimation — identifies abnormal human movements in footage. Developed during the UNDP × Samsung Innovation Campus bootcamp. Dataset: Kaggle ↗
End-to-end ML pipeline predicting customer churn using Gradient Boosting, Random Forest, and SVC. Includes KMeans segmentation, Evidently AI monitoring dashboards, and a live Streamlit app.
View on GitHub →Deep learning model detecting fraudulent transactions with 97% accuracy across 29 anonymized features. Combines Keras neural networks with Logistic Regression and SVC; handles severe class imbalance via undersampling.
View on GitHub →