Building Agentic AI Digital Twins with Kirill Eremenko

Understanding Agentic AI

• 🧠 Agentic AI is conceptualized as an LLM brain with access to memory, Retrieval-Augmented Generation (RAG), and tools, enabling it to perform actions with a degree of agency.
• 💡 A key insight is that agents do not directly call tools; instead, the LLM within the agent communicates requests to the system running it, which then executes the tool calls.
• 💬 All interactions between the LLM and the system, including tool requests and responses, occur through text-based prompts and replies.

Pro-Tips for Agentic AI Development

• 🛠️ Routing tool calls to the appropriate functions is best handled by the external code system rather than the LLM itself, especially in complex scenarios with multiple tools.
• ⚖️ The system should intelligently decide which tool to use (e.g., a calculator for math, Gmail for email queries) to ensure accurate responses.

Digital Twin AI Agent Creation

• 👤 A practical application demonstrated is the creation of a digital twin AI agent, acting as an online alter ego.
• 📚 This agent can be augmented with personal data such as LinkedIn profiles, resumes, blogs, GitHub repositories, and even bootcamp transcripts.
• 🗣️ The digital twin is capable of answering interview-style questions based on the provided biographical information, simulating a candidate's experience and knowledge.

Real-World Application and Success

• 🚀 One participant successfully used their built digital twin agent during a job interview, showcasing its capabilities to the interviewer.
• 💬 The participant also explained the technical implementation of the agent, including system prompts, RAG, and memory usage, impressing the interviewer.
• ✅ This led to the participant landing an AI engineering job, highlighting the practical value of building and deploying such agents.
• 📈 The agent can be further enhanced with features like SMS notifications for usage and interactive interfaces.