Home
Student Projects
Aegis Framework: A Multi-agent Ai Platform For Biomedical Research

Aegis Framework: A Multi-Agent AI Platform for Biomedical Research

GSK-IE Biopharma and AI Gateway Specialization Project

Aegis Framework is an AI-powered research support platform developed through the GSK-IE Biopharma and AI Gateway Specialization to assist researchers tackling antimicrobial resistance. Using a multi-agent architecture and retrieval-augmented generation, the platform analyzes patient case studies, synthesizes scientific evidence, and delivers structured recommendations to support evidence-based decision-making.

"The best way to understand emerging technologies is to build with them. Theory gives you foundations, but real learning happens when you apply those tools to meaningful problems."

Project Overview

What was your project about?

Our project, called Aegis Framework, was developed as part of the GSK-IE Biopharma and AI Gateway Specialization. It is an AI-powered platform designed to support researchers working on antimicrobial resistance (AMR), specifically tuberculosis as our initial focus area. The system uses multiple AI agents with different specialized roles and priorities to analyze patient case studies, conduct evidence-based research, and provide structured recommendations to researchers through a web application currently still in development.

Purpose of the project

What was the main purpose behind choosing this project, and what impact or real-world problem were you aiming to address?

Antimicrobial resistance is becoming a major global health challenge due to the overuse and misuse of antibiotics. Researchers and healthcare professionals often need to evaluate many complex variables before making treatment-related decisions, which can be time-consuming and resource-intensive. Our goal with Aegis Framework was to explore how AI could help reduce that workload by providing faster, evidence-backed research support while still keeping human experts at the center of the decision-making process.

Technical Details

What technical approaches and tools did you use, and why did you choose them?

We built the platform using React, TypeScript, and HTML for the frontend, while integrating APIs from OpenAI, Anthropic, Google Gemini, and Ollama to leverage the strengths and diversity of different large language models. We also implemented a retrieval-augmented generation (RAG) system with a database that allows researchers to upload their own scientific documentation. We chose a multi-agent architecture because it enabled different AI agents to evaluate cases from different perspectives and cross-check each other’s reasoning, helping improve consistency and reliability.

What knowledge, tools, or theories from your program courses did you apply to this project?

This project combined both technical and business knowledge from our programs. We applied concepts from data analytics, AI systems, programming, and research methodology, while also using project management and teamwork skills from the business side of our studies. One of the most important lessons from the program was learning how to approach ambiguous and interdisciplinary problems in a structured way.

Challenges and Solutions

Were there any significant challenges during this experience, and how did you overcome them? Did what you learn in the program help you along the way?

One of our biggest challenges was building a system that researchers could trust, especially in a healthcare-related environment where AI hallucinations can have serious consequences. To address this, we constrained our sources to scientific publications, required all AI-generated claims to include references, and implemented multi-agent validation processes to surface inconsistencies. Another challenge was working in a highly specialized biomedical field despite coming from more technical and business-oriented backgrounds. Our mentors played a critical role in helping us understand the scientific perspective, while our coursework helped us adapt quickly and structure the project effectively under tight time constraints.

Collaboration and Teamwork

How did teamwork contribute to the success or progress of your project?

Teamwork was essential because the project was both technically and conceptually ambitious. We divided responsibilities based on our strengths: Sibylle and Juan focused mainly on the technical implementation and data-related aspects, while Carlos worked extensively on AI agent personas and prompt engineering. We also held weekly meetings to maintain accountability, track progress, and continuously improve the platform. That structure helped us stay consistent and avoid last-minute development.

Learning and Takeaways

What key lessons and skills (soft and hard) have you acquired or strengthened through this experience?

Technically, we strengthened our skills in AI systems, API integration, prompt engineering, frontend development, and working with retrieval-augmented generation architectures. On the soft skills side, we improved significantly in teamwork, interdisciplinary communication, adaptability, and managing complex projects under uncertainty. This experience also taught us that the best way to truly understand emerging technologies is by actively building with them.

Future Development

Do you have plans for further development or improvement of your project in the future?

Yes. Over the summer, we plan to continue developing the platform to make it more robust, scalable, and reliable. One of our main priorities is further reducing hallucinations and improving the factual consistency of the AI-generated recommendations. We also want to expand the framework beyond tuberculosis into other diseases and continue updating the models as AI technology evolves. Additionally, we are hoping to publish our work together with our mentor and continue testing the platform in more realistic research environments.

ADVICE

What advice would you give to other students who want to get involved in this type of initiatives and tech immersions?

Our advice would be to not wait until you feel completely ready before starting. The best way to learn about new technologies is to actively work with them and apply them to meaningful problems. Interdisciplinary projects can feel intimidating at first, but they are also some of the most rewarding experiences because they push you to learn quickly, collaborate effectively, and think beyond your comfort zone.