#59: The Agentic AI Era, Smolagents, and a “Gatekeeper” Agent Prototype

Author(s): Towards AI Editorial Team

Originally published on Towards AI.

Good morning, AI enthusiasts! As you already know, we recently launched our 8-hour Generative AI Primer course, a programming language-agnostic 1-day LLM Bootcamp designed for developers like you.

We also have a special discount for our community members (yes, that’s you!). Use code towardsai_8hour to get 15% off on the course. So don’t wait, learn to make the most of LLMs before the next big AI update drops. Start here!

As always, we have practical tutorials, collaboration opportunities, a fun prototype from the community, and a lot more!

P.S. If you are already interested in our LLM developer course, use the code towardsai_8hour on our bundle offering, which includes this new course and our more in-depth companion course, ‘From Beginner to Advanced LLM Developer’. Check out the bundle offering here!

What’s AI Weekly

This week in What’s AI, I’m diving into the world of APIs — what they are, why you might need one, and what deployment options are available. When we talk about building powerful machine learning solutions, like large language models or retrieval-augmented generation, one key element that often flies under the radar is how to connect all the data and models and deploy them in a real product. This is where APIs come in. Read the complete article here or watch the video on YouTube!

— Louis-François Bouchard, Towards AI Co-founder & Head of Community

Learn AI Together Community section!

Featured Community post from the Discord

Malus_aiiola has built an AI voice “Gatekeeper” agent that handles incoming calls for busy CEOs. In this video, he breaks down how you can benefit from an AI voice “Gatekeeper,” which will answer for you and record the information of the caller. You can also get the blueprint and prompts to make it yourself. Reach out in the thread if you want to build something like this.

AI poll of the week!

Is Devin promising? Is it conceptually good, or do you see yourself adopting it? Tell us in the thread!

Collaboration Opportunities

The Learn AI Together Discord community is flooding with collaboration opportunities. If you are excited to dive into applied AI, want a study partner, or even want to find a partner for your passion project, join the collaboration channel! Keep an eye on this section, too — we share cool opportunities every week!

1. Mr_oxo is looking for people to collaborate with on Computer Vision projects as accountability partners and problem-solving buddies. If you’re passionate about computer vision and want to level up your skills while working on projects, connect in the thread!

2. Pulkitplays is building a Next Ball Score Prediction Model and needs a project partner to solve the problem statement. If this sounds interesting, reach out in the thread!

Meme of the week!

Meme shared by rucha8062

TAI Curated section

Article of the week

Graph Neural Networks: Unlocking the Power of Relationships in Predictions By Shenggang Li

This article explores Graph Neural Networks (GNNs), focusing on their ability to analyze connected data. It explains how GNNs interpret nodes and edges, using examples like cities connected by roads. The article then delves into three GNN types: Convolutional GNNs for economic forecasting, Graph Attention Networks for feature selection, and Temporal GNNs for stock prediction. Code examples and results demonstrate each GNN’s application, including predicting economic indicators, identifying key dataset features, and forecasting stock trends. It highlights GNNs’ broader potential across diverse fields like healthcare, traffic management, and information retrieval with LLMs.

Our must-read articles

1. Smolagents + Web Scraper + DeepSeek V3 Python = Powerful AI Research Agent By Gao Dalie (高達烈)

This article provides a tutorial on creating a multi-agent chatbot using Smolagents, a Python library for building AI agents, combined with web scraping and the DeepSeek V3 language model. The author highlights Smolagents’ simplicity, requiring minimal code to create agents capable of complex tasks, including Agent-Retrieval-Generation systems. The DeepSeek V3 model was chosen for its cost-effectiveness and performance, which are comparable to GPT-4 and Claude. It concludes by emphasizing Smolagents’ efficiency and ease of use for developing sophisticated AI agents.

2. Building an On-Premise Document Intelligence Stack with Docling, Ollama, Phi-4 | ExtractThinker By Júlio Almeida

This article details building an on-premise document intelligence solution using open-source tools. It addresses the data privacy concerns of financial institutions by leveraging local language models (LLMs) like Phi-4 via Ollama, combined with ExtractThinker for orchestrating document processing and Docling/MarkItDown for handling document loading and OCR. It guides readers through choosing appropriate LLMs (text vs. vision-based), document parsing libraries, and local deployment solutions. It also provides strategies for managing limited context windows in local models, including lazy splitting for large documents and pagination for partial responses. Code examples demonstrate integrating these components for a complete document extraction pipeline, offering a secure and compliant solution for sensitive data processing. Finally, it discusses PII masking for cloud-based LLM usage when local deployment isn’t feasible.

3. Building Multimodal RAG Application #8: Putting it All Together! Building Multimodal RAG Application By Youssef Hosni

This article presents a comprehensive guide to building a multimodal Retrieval-Augmented Generation (RAG) application, culminating the eight-part series. It integrates modules for preprocessing multimodal data, retrieving information using a LanceDB vector store, and generating responses with Large Vision Language Models (LVLMs) via a PredictionGuard client. It details setting up the environment, processing data (including image-text embedding), building the retrieval module, implementing LVLM inference, and processing prompts. Finally, it demonstrates combining these modules into a functional multimodal RAG system using LangChain, enabling complex queries against diverse data types like text and video.

4. The Agentic AI Era: A Primer By Kaush B

This article provides a primer on the Agentic AI Era, exploring the evolution of AI from automation tools to autonomous agents. It defines AI agents, categorizing them by type and architectural topology, and outlines their characteristics and developmental stages within an enterprise context. It details various agentic workflows, including prompt chaining, routing, parallelization, orchestrator-workers, and evaluator-optimizer models. It also discusses four key agentic design patterns: reflection, tool use, planning, and multi-agent collaboration. Practical implementation guidelines, technical risks, and mitigation strategies are also addressed. It concludes by emphasizing the transformative potential of agentic AI while acknowledging the ethical and societal challenges that require careful consideration for responsible deployment.

If you are interested in publishing with Towards AI, check our guidelines and sign up. We will publish your work to our network if it meets our editorial policies and standards.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI