Teknofest Action-Based Turkish Large Language Model Competition

As part of Türkiye’s National Artificial Intelligence Strategy, this technology competition aims to strengthen the Turkish large language model (LLM) ecosystem. Organized by TÜBİTAK, the Türkiye Technology Team (T3) Foundation, and the Ministry of Industry and Technology, the competition concludes within the TEKNOFEST Istanbul final events.

The competition is designed to cultivate skilled human resources, contribute to national strategies, and encourage innovative solutions in Turkish LLM development. It seeks to enhance the expertise of researchers, students, entrepreneurs, and industry professionals, positioning Türkiye as a representative player on the international stage in this field.

Participants are expected to fine-tune open-source base models with up to 10 billion parameters using sector-specific datasets, and enhance these models by adding software layers such as AI agents to perform tasks in vertical domains including Education, Tourism, or Public Technologies—all in Turkish.

The competition is open exclusively to participants within Türkiye. Undergraduate and graduate students, graduates, researchers, entrepreneurs, and corporate or R&D teams may apply. Teams consist of 3 to 6 members (including an advisor), and each team must designate a captain and a communication officer.

• 20 June – Preliminary Project Evaluation Report and final application deadline
• 6 July – Online Exam
• 13 July – Announcement of teams advancing to the second stage
• August – Leaderboard Platform closure
• September – Final report submission, finals, and award ceremony

1. Preliminary Evaluation: Technical competence is assessed through a project report and an online exam.
2. Application Development:
3. 1. Participants may use the provided high-performance infrastructure or their own resources for model fine-tuning.
   2. Checkpoints on the Leaderboard Platform are evaluated based on accuracy, speed, and reliability metrics.
   3. An AI agent is developed for the selected vertical domain using the fine-tuned model.
   4. The process is completed with submission of the final report, a live demonstration of the working application, and jury evaluation.

Projects must not include proprietary, closed-source, or paid dependencies. Base models must be open-source; distilled or quantized versions are acceptable. Participants are responsible for ensuring the licensing and ethical compliance of their fine-tuning datasets.

The final score is calculated as a weighted sum of four components:

• Model Fine-tuning (35%)
• Dataset Quality (15%)
• Final Report (20%)
• Jury Presentation (30%)

Teams placing in the student category receive:

• First Place – 150,000 ₺ + 200 hours of H100 GPU access
• Second Place – 120,000 ₺ + 150 hours of H100 GPU access
• Third Place – 100,000 ₺ + 100 hours of H100 GPU access
• Additionally, the “Young Talent Award” is granted based on original contributions and team collaboration.