Kurs
Building AI Agents with Google ADK
1 Std.
3K
AI agents are advanced, goal-driven systems that perceive context and execute multi-step workflows with minimal supervision, unlike basic chatbots. They can call APIs, operate software, query data, and leverage memory and feedback loops.
These agents have become a core part of the AI ecosystem, and learning about them will help you launch your AI career or improve your current skills. The best way to understand any new concept is through building practical applications.
In this article, we will review 10 AI agent projects useful for every skill level:
- Beginners: Build quickly with low-code tools like Langflow, Flowise, and Make AI.
- Intermediate: Use frameworks like LangGraph, Mistral Agents, and Qwen-Agent to create custom agents.
- Advanced: Design multi-agent systems with Haystack, ADK, and CrewAI.
Beginner AI Agent Projects
Easy projects use GUI or low‑code agent tools. You can visually drag, drop, and connect components, add your LLM API key, and run the pipeline. These builders let you prototype stateful agents, integrate data sources and APIs, and chain prompts and tool calls without boilerplate code.
1. Language Tutor With Langflow
Language Tutor with Langflow is a small agentic system that generates bite‑sized reading passages tailored to the learner’s current vocabulary. It runs on Langflow with a Postgres database (via Docker) and uses psycopg2 to read/write words.
You upload your vocabulary from a CSV, add new words through a tool in chat, and a story tool pulls your saved words to prompt an LLM to write a story in your chosen language. The main agent routes your request, “add word” or “make a story”, and returns the result.
Guide: Langflow: A Guide With Demo Project
2. Data Analyst AI Agent With Flowise
Data Analyst AI Agent with Flowise is a workflow that lets you ask questions about a database and get answers with the exact SQL used.
You connect Flowise to a SingleStore database, add a custom code block to read table schema, and feed that into a prompt that asks an LLM (via an LLM Chain with OpenAI) to generate a SQL query.
The query is stored, cleaned, executed via another custom code block, and the results plus the query are formatted by a final prompt and LLM Chain.
Guide: Flowise: A Guide With Demo Project
3. Automating Customer Service With Make AI
Customer Service AI Agent with Make automatically replies to rental inquiries submitted through a Tally form.
When someone fills out a Tally form, Make AI grabs the rental details from a Google Doc. Then, an OpenAI module writes a reply using those details and the person’s question. Finally, the email module sends the answer straight to the email address from the form.
Guide: Make AI: A Guide With Practical Examples
Intermediate AI Agent Projects
Intermediate AI Agent projects focus on building complete workflows using modern agentic frameworks and APIs. At this stage, you are not just connecting APIs, you are also designing simple UIs so users can interact with the agents directly.
4. Nutrition Coach With Mistral Agents
Nutrition Coach with Mistral Agents is an AI agent project that logs your meals, estimates calories, and suggests a healthy next meal, with an image.
The app includes Mistral’s Agents API, which features a web search agent for obtaining calorie estimates. It also has a fallback estimator, a logger to record meals, calorie counts, and timestamps, as well as an image-generation agent to visualize the suggested dish.
Users can enter their meal and preferences, and the app will search for or estimate calorie content, log the entry, suggest a follow-up meal, and display an auto-generated image of the dish, along with a clear summary of the tools used.
Guide: Mistral Agents API: A Guide With Demo Project
5. Deep Research Assistant With Jan-v1
The AI Agent project Deep Research Assistant with Jan‑v1 is an app that turns a topic into a polished research report using local Jan‑v1 inference, async web search, and strict report formatting. It generates smart queries, pulls sources, and synthesizes a clean, professional write‑up.
It uses Streamlit for the UI, llama‑cpp to run a Jan‑v1 GGUF model locally, Serper for web search, and a set of helper functions for sectioning and cleanup.
Users enter a topic, choose depth, focus, timeframe, and format; the app generates queries, runs async searches, compiles notes, and has Jan‑v1 produce a structured report, then displays sources and lets you export TXT/JSON. Progress bars track steps, and session state avoids reloading the model between runs.
Guide: Jan-V1: A Guide With Demo Project
6. Real-Time Web Summarizer Extension with Qwen-Agent
Real-Time Web Summarizer extension is a Chrome add-on that captures the visible text of any page and streams a clean, concise summary in real time, powered locally by Qwen3 via Ollama and a FastAPI backend.
Users can click “Summarize” in the pop-up, as shown below. The extension retrieves the page text, sends it to http://127.0.0.1:7864/summarize_stream_status, and displays the streaming response.
To set it up, you need to pull the Qwen model using Ollama, start the FastAPI server, load the unpacked extension, and then you can get instant summaries on any page, presented in a clear, editor-style format.
Guide: Qwen-Agent: A Guide With Demo Project
7. Real-time Analytics with LangGraph
Real-time Analytics is a LangGraph-powered assistant that answers questions, searches the web, and executes Python code. It combines Mistral Medium 3 for reasoning, Tavily for web search, and a Python REPL for running code based on user prompts.
When users ask a question, the agent determines whether to perform a search, execute code, or use both methods. It then provides the final answer along with information about the tools used. The setup process includes adding the Tavily API key, installing necessary packages, initializing the LLM and tools, creating the LangGraph agent, and invoking it with user messages.
Guide: Mistral Medium 3 Tutorial: Building Agentic Applications
Advanced AI Agent Projects
Advanced AI Agent projects bring together multiple agents working with multiple tools in a single workflow. These projects make use of powerful agentic frameworks such as Haystack, ADK, and CrewAI, allowing you to design complex, collaborative systems where agents can coordinate, specialize, and tackle more sophisticated tasks.
8. RAG and Web AI Agents with Haystack AI
Haystack Agentic RAG and Web Access is an assistant that answers questions using a private knowledge base and, when needed, live web results. It routes queries to retrieval-augmented generation first, and switches to web search for fresh, real-time info.
It uses Haystack pipelines and agents, an in-memory document store with OpenAI embeddings, a custom RAG tool, and a custom Tavily web search tool wrapped as ComponentTool. The agent is powered by GPT 4.1 Mini with a system prompt that guides tool selection.
Users can pose questions, and the agent retrieves information from the knowledge base or conducts a Tavily search for current topics. It then provides an answer along with the tools used.
Guide: Haystack AI Tutorial: Building Agentic Workflows
9. Travel Planner with ADK and A2A
Travel Planner with ADK and A2A is a comprehensive multi-agent application designed to plan trips from start to finish. Users simply input their destination, travel dates, and budget. Specialized agents then recommend flights, accommodations, and activities. A coordinating host agent organizes all the components, while a Streamlit user interface displays the complete itinerary.
Users fill the form in Streamlit, the host agent sends the payload to all three agents, merges their JSON responses, and returns a structured plan with flights, hotels, and activities.
Guide: Agent Development Kit (ADK): A Guide With Demo Project
10. Agentic RAG with CrewAI
Agentic RAG pipeline is a query-routing system that answers with grounded context from local PDFs or, when needed, the live web.
It uses FAISS over PDF chunks, Groq for fast LLM responses, and a crewAI web workflow for external context. A router prompt determines “Yes/No” on whether local data is enough, and utilities handle retrieval, web scraping, and synthesis.
Users ask a question, and the router checks the PDF context. If the answer is "Yes," the system retrieves the top matches from the vector database. If the answer is "No," it searches the web, combines the gathered information, and the language model formulates the final response.
Guide: Agentic RAG: Step-by-Step Tutorial With Demo Project
Final Thoughts
Building your own AI projects from scratch is one of the best ways to boost your career. It helps you go beyond theory and gain practical skills. You will learn how to define a problem, connect tools and agents, validate results, create a simple user interface, and improve based on real user feedback.
Each project you complete adds to your portfolio, which you can share on platforms like GitHub or Hugging Face, or as a live demo. This shows potential employers that you can design and deploy reliable systems, not just use existing models. It demonstrates your understanding of product development and engineering, making you more attractive to hiring managers in fields like applied machine learning and AI engineering.
I recommend starting with small projects and aim to release them frequently. Let your work speak for itself, each project tells the story of your growth and expertise in the AI field. If you’re new to agentic AI, I also recommend taking the Introduction to AI Agents course and checking our our AI Agents Cheat Sheet.
