80 Students Build Autonomous Coding Agents at Tec de Monterrey

Last week marked the conclusion of one of the most ambitious capstone projects we’ve undertaken at Tecnológico de Monterrey. Over the course of a semester, 80 students built autonomous coding agents by fine-tuning small language models on specific code repositories.

The Challenge

Students were tasked with creating AI assistants capable of reasoning about code problems. This wasn’t just about using existing tools—they had to:

• Apply compiler techniques to filter and curate datasets from large code corpora
• Fine-tune language models on repository-specific codebases
• Deploy cloud-based APIs for their AI assistants
• Build agents with genuine reasoning capabilities

Technical Achievements

The teams demonstrated impressive technical sophistication:

• Custom dataset curation pipelines
• Model fine-tuning with LoRA and other efficient techniques
• Cloud deployment on Modal infrastructure
• Integration with real-world development workflows

Industry Partnership

This project wouldn’t have been possible without our partners:

• C3 AI - Training and mentorship support
• Modal Labs - Cloud infrastructure for deployment
• Faculty support from Victor Rodriguez, Daniel Flores Araiza, Rodolfo Castelló, and Liliana Barbosa

Several teams were invited to present their findings in Redwood City, demonstrating the industry relevance of their work.

Why This Matters

Technology education must stay current with industry demands. Academic programs should bridge the gap between classical computer science fundamentals and emerging agentic AI applications. This capstone shows that students can master cutting-edge AI techniques while building practical, deployable solutions.

The future of software development is increasingly AI-augmented, and our graduates are ready to lead that transformation.