2-Hour Agent Projects

Option 1: Wikipedia + DuckDuckGo

Project: "Research Assistant - Compare Sources"

• Agent researches a topic using Wikipedia AND web search using DuckDuckGo

• Compares/contrasts information from both sources

• Generates brief report

Why this works:

• Zero setup friction (no API keys for DDG)

• Two tools reinforce same concept (information retrieval)

• Shows value of multiple information sources

• Students learn tool combination patterns

Learning outcomes:

• Using multiple tools in sequence

• Information synthesis from different sources

• Basic agent reasoning about which tool to use when

Option 2: OpenWeatherMap

Project: "Smart Travel Planner"

• Agent takes destination city and travel dates

• Fetches weather forecast

• Generates packing list and activity recommendations based on conditions

Why this works:

• API key practice (important real-world skill)

• Structured data handling (JSON parsing)

• Clear cause-effect (weather → recommendations)

• Real-world utility (students actually might use this!)

Learning outcomes:

• API authentication patterns

• Working with structured API responses

• Conditional reasoning based on external data

Option 3: YouTube Transcript

Project: "Educational Video Analyzer"

• Student provides educational video URL

• Agent extracts transcript

• Generates: summary, key concepts, quiz questions

• See below for more details and starter code!

Why this works:

• Reinforces custom tool creation (builds on calculator exercise)

• High student engagement (analyze videos they actually watch)

• Multiple LLM calls in sequence (fetch → summarize → generate questions)

• Visible value (helps with studying!)

Learning outcomes:

• Custom tool definition patterns

• Multi-step agent workflows

• Content transformation pipelines

Implementation Template (Works for Any of These)

Here's a 2-hour timeline that works for all three options:

Hour 1: Setup & Basic Tool

• 0-15 min: Install packages, get API keys (if needed)

• 15-45 min: Define custom tool OR use pre-built tool

• 45-60 min: Test tool in isolation (not in agent yet)

Hour 2: Agent Integration

• 60-90 min: Integrate tool with simple ReAct agent

• 90-110 min: Test with various prompts, debug

• 110-120 min: Optional extension or demo preparation

Detailed Recommendation: Start with YouTube

If I had to pick one for your next project after calculator, I'd choose YouTube Transcript.

Reasoning:

1. Builds directly on calculator pattern - they just wrote a custom tool, now they write another

2. No API key friction - can start coding immediately

3. High engagement - students love analyzing YouTube content

4. Clear progression - shows how custom tools enable new capabilities

5. Teaches real skill - most LangChain tools will be custom in production

Starter code for students:

python

​x
from youtube_transcript_api import YouTubeTranscriptApi
from langchain.tools import tool
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

@tool
def get_youtube_transcript(video_id: str) -> str:
    """Fetch the transcript of a YouTube video by video ID."""
    transcript = YouTubeTranscriptApi.get_transcript(video_id)
    return " ".join([entry['text'] for entry in transcript])

# Create agent with tool
llm = ChatOpenAI(model="gpt-4o-mini")
agent = create_react_agent(llm, [get_youtube_transcript])

# Test it
result = agent.invoke({
    "messages": [("user", "Get the transcript for video dQw4w9WgXcQ and summarize it")]
})

Extensions for faster students:

1. Add summary formatting (bullet points, key quotes)

2. Extract chapter timestamps

3. Answer specific questions about video content

4. Compare transcripts from multiple videos

This gives you a solid 2-hour project that reinforces custom tool creation while introducing practical information retrieval patterns.