Try it out on Hugging Face Spaces or click the button in the bottom right corner of the screen.
The retrieval side of this agent builds on the RAG pipeline I evaluated in depth in my Natural Language Processing project, where the same retrieve-then-generate approach was measured against a plain language model.
Project Overview
The Interactive Portfolio Agent is an autonomous system capable of answering detailed questions about my professional background, projects, and skills. It acts as an interactive version of my CV, leveraging LangGraph for orchestration and Retrieval Augmented Generation (RAG) to provide accurate, context-aware responses based on my actual portfolio data.
Key Features
- Autonomous Reasoning: Uses LangGraph to plan multi-step actions and execute tool calls efficiently.
- RAG & Tool Integration:
- Portfolio Retrieval: Searches and retrieves details about specific projects from markdown files.
- Structured Data Access: Queries structured YAML data for Experience, Education, and Certifications.
- Persistent Memory: Maintains conversation context to support follow-up questions.
- Robustness & Security: Includes rate limiting and token usage tracking.
- Interactive UI: Features a user-friendly Gradio interface and exposing a FastAPI endpoint.
Architecture
The agent is built with a focus on modularity and deployment:
- LangGraph: Manages the agent’s state and decision-making process.
- ChromaDB: Stores the vector embeddings for the RAG system.
- FastAPI: Serves the agent as a REST API.
- Docker: Ensures consistent deployment across environments, including Hugging Face Spaces.