AdvancedИнженерияgemini

Gemini in Practice: Real Google Examples

8 advanced examples from the Google team: ReAct agent, search grounding, PDF entity extraction, function calling chatbot, anomaly detection with embeddings, and GitHub issue analysis.

4modules

8lessons

240 mintotal time

Developers applying Gemini to real projectsaudience

Module 1

Agents and Tools

ReAct agent, search grounding, automatic function calling — patterns for building agentic systems with Gemini.

ReAct Agent: Reasoning + Acting

Implement the ReAct (Reasoning + Acting) pattern with Gemini 2.0 Flash and the Wikipedia API. The model generates Thought→Action→Observation chains to answer complex questions.

30 min

Search Grounding: Real-Time Data in Responses

Connect Google Search Tool to Gemini 2.0+ to generate current business reports with live data. Understand the difference between search grounding and the new search tool.

30 min

Barista Bot: Agent with Automatic Function Calling

Build a fully functional coffee shop ordering agent using automatic function calling. The SDK generates schemas from docstrings, calls functions, and returns results to the model automatically.

30 min

Module 2

Structured Data Extraction

Entity extraction from text and PDF documents using Pydantic schemas and structured output.

Entity Extraction from Text

Use Gemini to extract structured entities from arbitrary text: addresses, phone numbers, URLs, street names, and transportation routes.

30 min

Structured Output from PDFs: Invoices and Forms

Upload PDFs via the File API, define Pydantic schemas for invoices and handwritten forms, and receive structured JSON with guaranteed type validation.

30 min

Module 3

Embeddings: Search and Anomaly Detection

Practical applications of embeddings: detecting anomalies in data and re-ranking search results by semantic similarity.

Anomaly Detection with Embeddings

Use text-embedding-004 to vectorize 20 Newsgroups texts, visualize clusters with t-SNE, and detect outliers by distance from the cluster centroid.

30 min

Search Re-Ranking with Embeddings

Combine Gemini function calling with the Wikipedia API for search, then use embeddings and cosine similarity to re-rank found results by relevance.

30 min

Module 4

Data Analysis and Integrations

Analyzing GitHub issues via function calling and PyGithub, automating tasks that interact with external APIs.

GitHub Issue Analyzer: Repository Analysis with Gemini

Use PyGithub to fetch repository issue data and Gemini with function calling for interactive analysis: label filtering, stale issues, and statistics.

30 min