Welcome to My World

About Me

Hi! I’m Ako, a Data Scientist, AI enthusiast, and Climate Researcher. I specialize in blending Artificial Intelligence and Climate Science.

Interests:

• Fine-tuning Large Language Models (LLMs)
• Physics-Informed Deep Learning
• Kolmogorov-Arnold Representation Theorem
• Chess Enthusiast (Expert Level)

Get in Touch:

Check out my work on GitHub.

Projects

• Modular LLM Architecture : A multi-agent system leveraging LLMs for scalable workflows.

Multi-Agent LLM Framework at NBC

As part of my role at NBC, I worked on designing and implementing a cutting-edge multi-agent framework to leverage Large Language Models (LLMs) for handling complex user queries dynamically. The project aimed to enhance the efficiency and accuracy of conversational agents by orchestrating multiple agents for specialized tasks.

Background

With the rise of LLMs, managing complex workflows with diverse data sources became critical for NBC's data operations. Traditional single-agent approaches were insufficient for handling multi-step tasks, such as retrieving knowledge, summarizing data, and executing SQL queries.

Approach

• Designed a multi-agent architecture where each agent was responsible for a specific role, such as knowledge retrieval, SQL execution, or summarization.
• Leveraged frameworks like LangChain and RAG (Retrieval-Augmented Generation) for dynamic agent orchestration.
• Integrated multiple LLMs and APIs to ensure robust responses across varied tasks.
• Developed fallback mechanisms to handle failures gracefully and improve user experience.

Key Achievements

• Improved query resolution time by 30% through task-specific agents.
• Increased accuracy of responses by integrating fine-tuned retrieval models with the LLM workflow.
• Successfully deployed the system to support NBC's internal tools, handling thousands of queries daily.

Takeaways

This project underscored the importance of modularity and collaboration between agents in scaling AI systems. Future iterations will explore enhanced agent communication and self-learning mechanisms.