From NLP research to real-time streaming systems and deployed ML pipelines.
NLP · LLM · Streamlit
RAG Document Q&A System
Multilingual (Azerbaijani / Turkish / English) document Q&A system powered by Retrieval-Augmented Generation. Upload any PDF, ask questions, get answers with source page references.
FAISS vector store
Phi-3 / Mistral LLM
3-layer architecture
PythonLangChainFAISS
Sentence-TransformersStreamlitPyMuPDF
Big Data · Streaming · Real-time
Kafka Real-Time Data Streaming
Apache Kafka-based real-time data streaming pipeline. Producer–consumer architecture with distributed topic management, fault tolerance, and scalable big data processing.
Pub-Sub model
Real-time processing
Apache KafkaPython
Producer/ConsumerBig Data
Deep Learning · Computer Vision · PyTorch
Apple Leaf Disease Detection (CNN)
Deep learning image classifier for apple leaf disease detection across 4 disease classes using CNN in PyTorch. Outperforms Random Forest baseline by 8.7% with 97.71% precision.
97.71% precision
97.69% recall
4,000+ images
PyTorchCNNScikit-learn
MatplotlibPlantVillage
NLP · Sentiment Analysis · Web Scraping
KFC Reviews Sentiment Analysis
Scraped 500+ Trustpilot reviews using Selenium, conducted sentiment analysis with TextBlob, and built 3 classification models. Detected mismatches between written tone and star ratings.
91.7% LR accuracy
0.97 F1-score
0.61 Pearson r
PythonSeleniumTextBlob
TF-IDFScikit-learnSeaborn
Backend · REST API · DevOps
Todo API — FastAPI + PostgreSQL + Docker
Production-ready REST API built with FastAPI, PostgreSQL, and Docker. Full CRUD operations, Swagger UI docs, containerized deployment with docker-compose, and clean SQLAlchemy ORM architecture.
7 endpoints
Dockerized deploy
FastAPIPostgreSQLDocker
SQLAlchemyPythonPydantic
AWS · Snowflake · NLP · Analytics
Conversational Analytics — Regiocast
End-to-end NLP pipeline processing sales queries in German natural language. 3-layer Snowflake data model with LLM-generated SQL and self-serve Streamlit dashboard with CSV export.
100% query coverage
0 data exposures
AWS S3SnowflakeOpenAI API
PythonStreamlitNLP