I'm a Data Engineer with hands-on experience building scalable data pipelines and cloud-native solutions. I have architected comprehensive ETL workflows, real-time data ingestion systems, and LLM-powered solutions using AWS services.

MSc Thesis project implementing a serverless ticketing system using AWS services

• Tech Stack: AWS Lambda, Kinesis, DynamoDB, Step Functions, S3, AWS CDK, SNS, LangChain
• Description: An event-driven serverless ticketing system that processes support requests in real-time. Incoming tickets are ingested through Kinesis Data Streams and orchestrated by Step Functions. Each ticket undergoes sentiment analysis with Amazon Comprehend, followed by AI-generated responses using Bedrock LLMs via Lambda. Ticket metadata is stored in DynamoDB for fast retrieval, while complete records are archived in S3. SNS handles real-time notifications, and AWS Glue performs ETL operations to load data into Redshift for analytics. CloudWatch Alarms monitor the entire pipeline for failures, ensuring reliable ticket processing.

A dark-themed weather dashboard built with Streamlit and powered by the free Open-Meteo API

• Tech Stack: Python 3.13, Streamlit, Open-Meteo API, Folium, UV
• Description: A single-page weather dashboard that delivers real-time conditions for any searched city, including temperature with °C/°F toggle, humidity, wind speed, and wind direction. Uses Open-Meteo's geocoding and forecast APIs with response caching to reduce redundant calls. Features an interactive Folium map and a modular architecture separating services, models, and utilities.

LangGraph-based AI agent that generates high-quality prompts through automated self-critique and refinement cycles.

• Tech Stack: Python 3.13.5+, LangGraph, LangChain, Google Gemini API, Pydantic, LangSmith
• Description: Description: An agentic prompt engineering system implementing the reflection pattern where AI models examine and improve their own outputs iteratively. Uses a two-node LangGraph workflow (Generation + Reflection) with structured Pydantic outputs, comprehensive token tracking, and LangSmith observability. Each iteration incorporates feedback to progressively refine prompts, addressing edge cases, safety, and user experience. Demonstrated with a Wi-Fi troubleshooting chatbot that evolved from basic framework to production-ready prompt across 4 iterations. Ideal for technical documentation, customer support automation,requirements analysis, and complex content generation where quality matters more than speed.

• RAG pipeline