Atadan Icen

An intelligent tutoring SaaS that transforms rote memorization into question-driven learning using vision-based RAG. (Under development)

• Built a full-stack vision-based RAG SaaS as a solo founder
• Designed a scalable document ingestion and indexing pipeline
• Improved retrieval quality using agentic RAG techniques

An AI-powered travel planner using multi-agent orchestration to generate personalized travel plans.

• Implemented AI agent orchestration using CrewAI
• Deployed Ollama with an open-source LLM (OpenHermes)
• Built an interactive Streamlit UI

A personalized image generation system that creates realistic photos using user faces.

• Integrated Stable Diffusion with IP-Adapter for face conditioning
• Built a high-performance FastAPI backend
• Designed a user-friendly Streamlit interface

A data exploration app that simplifies NASA’s Near-Earth Object (NEO) API, enabling users to analyze asteroid flybys over extended date ranges.

• Bypassed the NASA NEO API’s 7-day limit to enable up to 31 days of asteroid data analysis
• Implemented sortable tables and CSV export for data-driven research and analysis
• Built an intuitive Streamlit interface for quick exploration

A lightweight currency converter and time-series visualization app using European Central Bank reference rates via the Frankfurter API.

• Implemented real-time currency conversion using ECB-backed exchange rates refreshed daily
• Built interactive time-series charts to analyze historical currency trends
• Enabled CSV export of historical exchange rate data for offline analysis

A machine learning system designed to detect potentially malicious forms such as credit card theft and spam.

• Developed a high-performance detection model using PyCaret and LightGBM
• Achieved 98% F1 score and 97% precision in identifying risky forms
• Reduced manual form review workload by 20%
• +2 more highlights

A semantic recommendation system that suggests similar form templates based on user intent.

• Developed a recommendation system using SBERT embeddings
• Identified semantically similar form templates with high accuracy
• Mined large-scale form templates using Selenium
• +1 more highlights

A real-time system for predicting future object trajectories using computer vision.

• Used YOLOv5 for object detection
• Applied Kalman Filters for trajectory forecasting
• Visualized predictions in real time with Streamlit