My Internship Journey: Lessons in Code, Collaboration, & Confidence

By Gavin Rosander

CANA Futures Program Intern

During my internship at CANA, I’ve had the opportunity to dive deep into a fast-moving, established project, evolving from core tasks to delivering system-wide impact. This journey has been defined by feature development, system validation, and the cultivation of foundational code discipline.

Technical Contributions & High-Impact Projects

My work focused on enhancing the intelligence and usability of a Supply Chain Assistant, a multi-agent system designed for complex data synthesis.

1. Dynamic Risk Report Tool

I built the initial tool to generate risk reports based on current context and database-stored graphs. To ensure the model never lacked necessary information, I expanded this tool with:

• Intelligent Retrieval: I added a web scraper to gather supplementary information if the model required more context.

• Client Delivery: The final output is a complete, dynamic report served directly to the user for immediate download.

2. Enhancing Application Quality & UX

To improve the experience for complex threads, I developed an intelligent chat history summarization feature that reduces clutter. Additionally, I am currently building a full authentication pipeline, which includes:

• Login & Registration: Developing the system for new application access.

• Security & Access Levels: Implementing role-based security to ensure appropriate access for different user types.

• Cosmos Integration: Integrating the system directly with Cosmos DB.

Validating the Multi-Agent System

A significant portion of my internship was dedicated to system validation and performance metrics. I assisted in writing performance metrics to test the reliability of agent tool calling and argument passing. One of my key results was achieving 100% accuracy in SPARQL-to-text fact-checking. I wrote and executed tests to measure how accurately the system translates technical SPARQL query results into natural language summaries. 

Growth Through Technical Discipline

My first professional experience with large-scale testing involved writing comprehensive unit tests for all MCP tools. This process instilled a vital lesson: clean, modular code is the foundation for testability. I learned that if a function is too confusing to test, it needs to be broken down—a realization that has made me a more disciplined coder.

The Power of the Environment & Mentorship

Beyond the code, the environment at CANA played a crucial role in my professional development. A cornerstone of this experience was "Trusting the Process" through dedicated mentorship. I found myself in a unique and lucky position to have a mentor I trusted from day one. This relationship drastically lowered the barrier to asking "stupid questions," which was critical for accelerating my learning and making me feel comfortable contributing to the team. Having a guide who encouraged self-guided strategies for system integration added immense value to my journey.

Strategy for Success: The "Sandbox"

Jumping into a large, established, and "moving" project was initially daunting. To build confidence, I developed a "sandbox" strategy, creating a local, simplified version of the core components. This hands-on approach allowed me to:

• Safely test ideas before making changes to the main project.

• Rapidly understand the core architecture.

• Develop a personalized method for learning that was essential for my style.

This combination of mentorship, technical discipline, and strategic learning allowed me to contribute meaningfully to the team while accelerating my personal growth.