AI Projects #9
AI Projects #9 Showcase will take place on Saturday, September 27 from 12.30 to 18:00, at Beykoz Kundura.
AI Projects is a 4-month research program where undergraduate, master's, and doctoral students collaborate to develop cutting-edge AI solutions. This intensive program provides a unique environment for participants to explore innovative ideas, conduct meaningful research, and build impactful projects.
Our ninth cohort brought together 25 passionate students who dedicated four months (May 30 - September 27) to developing 7 projects that address real-world challenges. From marine environmental monitoring to healthcare diagnostics, from cultural preservation to advanced NLP architectures, we are excited to share the results with you!
You can see the report of the 7th batch here and the 8th batch here.
Duration: 7 projects; 15 minutes presentation & 5 minutes Q&A each
Application Deadline: September 21 , 23.59.
Confirmation: by September 23, 23.59
We kindly ask you to apply for our exciting event due to limited space availability. You will be notified with a visitor form once your spot is confirmed!
Showcase Program
12.30 - 13.00 Welcoming & Onboarding
13.00 - 13.10 Ready to Showcase! (Pınar Yıldız, Program Manager, AI)
13.10 - 13.20 inzva’s Journey (Havva Yüksel, Program Lead, Algorithm)
13.20 - 13.30 AI Projects in Action (Zeynep Abalı, AI Projects Program Lead)
13.30 - 13.50 Spatiotemporal Forecasting of Marine Mucilage with Sentinel-2
and ERA5: A Computer Vision Approach
Simge Şenyüz, Ayşe Yeşil, Serkan Kızılırmak
13.50 - 14.10 Recipe Recommender using Computer Vision and LLMs
Kayra Kösoğlu, Mehmet Anıl Taysi, Simge Şenyüz
14.10 - 14.30 FolkBert : Robust Pose Estimation for Traditional Folk Dances
Melih Darcan, Ece Akdeniz, Muhammet Furkan Küpçü
14.30 - 15.30 Lunch Break
15.30 - 15.50 Hybrid SSM-Transformer Architectures for Long Context NLP
Kuzey Torlak, Anıl Dervişoğlu, Beyza Nur Deniz, Hüseyin Arda Aslan
15.50 - 16.10 DeTraffic: Deep Reinforcement Learning to De-Traffic Our Lives
Mehil Darcan, Volkan Bakır, Yusuf Koca
16.10 - 16.30 Coffee Break
16.30 - 16.50 Pediatric Appendicitis Detection on US Images
Hatice Karataş, Burak Emre Polat, Ege Uğur Amasya
16.50 - 17.10 Agent Based Task Oriented Dialogue System with Tool Calling Capabilities
Kerem Kosif, Eren Torlak, Umut Alkan, Yağmur Seda Tankutlu
17.10 - 17.30 Closing Remarks
THE PROJECTS
Spatiotemporal Forecasting of Marine Mucilage with Sentinel-2 and ERA5: A Computer Vision Approach
Simge Şenyüz, Ayşe Yeşil, Serkan Kızılırmak
This project leverages Sentinel-2 multispectral imagery (10–20 m resolution) and ERA5 climate variables (e.g., SST, wind, humidity) to detect and forecast marine mucilage in the Marmara Sea, the aim is comparing different algorithm performances and try to explain relation between climate variables and mucilage.
Recipe Recommender using Computer Vision and LLMs
Kayra Kösoğlu, Mehmet Anıl Taysi, Simge Şenyüz
This project develops an end-to-end pipeline that utilizes computer vision for the automatic detection and segmentation of ingredients from a user's fridge, combined with an LLM enhanced by a RAG framework to generate personalized and practical recipe suggestions. The primary goal is to create a system that accurately identifies food items and delivers highly relevant cooking recommendations.
FolkBert : Robust Pose Estimation for Traditional Folk Dances
Melih Darcan, Ece Akdeniz, Muhammet Furkan Küpçü
This project focuses on developing a robust pipeline for pose estimation in traditional folk dances. By introducing a newly curated dataset and employing fine-tuned models, it seeks to accurately capture intricate body movements under the challenges posed by traditional clothing and accessories. The proposed approach is expected to advance research in motion analysis, enhance educational tools for dance instruction, and support the preservation and systematic study of cultural heritage.
Hybrid SSM-Transformer Architectures for Long Context NLP
Kuzey Torlak, Anıl Dervişoğlu, Beyza Nur Deniz, Hüseyin Arda Aslan
This project explores a hybrid model combining State Space Models (SSMs) like Mamba with Transformer attention layers to improve long-context language modeling. The architecture aims to balance efficient long range memory with localized attention based reasoning. Different configurations (eg stacked, alternating) will be implemented and evaluated on long sequence benchmarks. Results will compare performance and speed against baseline SSM-only and transformer only models.
DeTraffic: Deep Reinforcement Learning to De-Traffic Our Lives
Mehil Darcan, Volkan Bakır, Yusuf Koca
This study presents a decision support system for appendicitis diagnosis using ultrasound imaging. The framework comprises three stages:
Localization: A YOLO-based detector identifies the appendix region in B-Mode ultrasound scans.
Segmentation: A U-Net model delineates the appendix contour and estimates its diameter.
Decision: Morphological features are combined with clinical and laboratory data, which are classified by an XGBoost model to produce the final diagnosis.
The system was trained and validated on static ultrasound slices, and will be further evaluated on dynamic video streams for real-time performance.
Pediatric Appendicitis Detection on US Images
Hatice Karataş, Burak Emre Polat, Ege Uğur Amasya
This project aims to create a better system for task-focused conversations to replace older models that are difficult to change and expensive to maintain. The new approach uses a team of specialized AI agents that work together, with each agent handling a key part of the conversation: understanding the user, tracking the chat's progress, deciding what to do next, and creating a response. This system will use large language models and can call on external tools to get things done, with the goal of building a conversational AI that is more flexible, easier to expand, and works better in real-world situations.
Agent Based Task Oriented Dialogue System with Tool Calling Capabilities
Kerem Kosif, Eren Torlak, Umut Alkan, Yağmur Seda Tankutlu
Customer service in the financial sector requires fast, accurate, and multilingual support, particularly for fraud detection and product inquiries. This project develops a modular, MCP-orchestrated chatbot that coordinates multiple specialized agents including Money, Fraud, Product, and Translation Agents to process real-time user queries. By integrating transaction history, account balances, credit information, and multilingual capabilities, the system leverages agent collaboration to handle diverse customer needs and ensure reliable interactions. The orchestration layer ensures smooth communication between agents, enabling context-aware and multi-agent dialogue management.
Our Space
How to get to inzva
By submitting this form you confirm that you read and understood inzva CODE OF CONDUCT and LETTER OF CONSENT.
A BEV Foundation project inzva is a non-profit hacker community organizing study and project groups as well as camps in the fields of AI and Algorithm; and gathering CS students, academics and professionals in Turkey.
Subscribe to our newsletter here and stay tuned and follow us on our social media accounts to have the most recent news about our upcoming events and programs!
All personal data you shared in this form will solely be used to determine the participants of the event and to inform the venue provider, and will be deleted as soon as we carry out our legal obligations. Therefore, by sending this form, you accept having your personal data processed and transferred to third parties that are providing services for the event (i.e. venue and transportation providers).