agentic_ai_project

Agentic AI Project with RAG Overview

This project implements an Agentic Retrieval-Augmented Generation (RAG) system capable of:

Searching the web for relevant URLs using DuckDuckGo.

Fetching and parsing HTML content from the URLs.

Summarizing content using LexRank.

Ranking content by semantic similarity to a user query using sentence embeddings.

Persisting content into a vector store (FAISS) for retrieval.

Generating a grounded answer using a large language model (Flan-T5) based on retrieved content.

This design allows the agent to answer user queries with contextually accurate information and optional citations. Project Architecture ┌───────────────────────────────┐ │ User Query │ └──────────────┬────────────────┘ │ ▼ ┌───────────────────┐ │ URL Search Agent │ │ (DuckDuckGo API) │ └─────────┬─────────┘ │ ▼ ┌───────────────────┐ │ Content Fetcher │ │ (requests + HTML) │ └─────────┬─────────┘ │ ▼ ┌───────────────────┐ │ Summarizer │ │ (LexRank) │ └─────────┬─────────┘ │ ▼ ┌───────────────────┐ │ Ranker (TF) │ │ Sentence Embeds │ │ Cosine Similarity │ └─────────┬─────────┘ │ ▼ ┌───────────────────┐ │ Vector Store (FAISS) │ └─────────┬─────────┘ │ ▼ ┌───────────────────┐ │ LLM Generator │ │ (Flan-T5) │ └─────────┬─────────┘ │ ▼ ┌───────────────────┐ │ Answer Output │ └───────────────────┘ 1- Setup Python environment - python -m venv agent_env -.\agent_env\Scripts\activate # Windows -source agent_env/bin/activate # macOS/Linux 2- Install dependencies pip install -r requirements.txt 3- Run the main agent: python main.py Enter your question: latest AI trends in 2026

=== Agentic AI Answer === AI agents are evolving from personal assistants to orchestrators of complex workflows. Research labs increasingly use AI to assist experiments, and industries adopt retrieval-based generation for accurate, context-aware responses.

=== Detailed Scores === URL: https://news.microsoft.co... Summary: Microsoft AI’s Diagnostic Orchestrator solved complex medical cases with 85.5% accuracy... Score: 0.6013

URL: https://www.ibm.com/think... Summary: LLMs are trained on specific domains and provide always-on assistance for enterprises... Score: 0.4046 4-Run all tests: pytest -v tests.py Includes unit tests for:

HTML cleaning

Summarization

URL ranking

Semantic similarity

Pipeline performance.

Key Features

RAG-enabled: Uses vector store to retrieve relevant chunks before generation.

Dynamic URL fetching: Works on live web content.

Summarization: Condenses long pages for efficiency.

Semantic ranking: Ensures top results are most relevant.

LLM-grounded generation: Produces accurate answers with context.