Inspiration
The idea for this project came from a simple observation: news is everywhere, but it’s often dry and repetitive. During hackathons, I wanted to create something that not only informs people but also entertains them. Humor is universal, so why not mix it with daily news? This inspired me to build an AI agent that fetches real-time news and blends it with jokes for an engaging reading experience.
I have kept everything at a very basic level yet very technically diverse by covering all the steps (Ingest & Index Data, Vector Search, Embedding Model, LLM Calls, External Tools, Multi-Step Flow). Anyone be it beginner or master can refer to my code and can build very complex and technical agents with very ease. My aim was to cover every aspect of the hackathon and help others by motivating them to start their Agentic AI journey.
What I Learned
- Vector databases (TiDB) for semantic search and efficient retrieval of jokes.
- HuggingFace embeddings to encode text for similarity comparisons.
- LangGraph workflows to orchestrate multiple tools dynamically.
- ChatGroq (Llama-3.1) for fast reasoning and natural language generation.
- Integrating web search APIs and structured AI tools to create a multi-step pipeline. I also improved skills in Python, API integrations, and building end-to-end AI workflows for real-world applications.
How I built it
Data Preparation: Collected jokes and stored them in TiDB with vector embeddings. Tool Creation: Developed two main tools - web_search for news and rag_search for jokes. Workflow Orchestration: Used LangGraph to coordinate the tools and LLM responses. Humorous Output: Designed prompts and a system message to instruct the LLM to blend news with jokes. Final Integration: Connected all components for smooth query handling and output generation.
Challenges I ran into
Balancing humor and information: Ensuring jokes complement news without overwhelming it. Embedding quality: Selecting embeddings that retrieved contextually relevant jokes. Tool orchestration: Preventing repeated queries and ensuring smooth flow between tools. API limitations: Handling inconsistent search results from DuckDuckGo efficiently.
Log in or sign up for Devpost to join the conversation.