1. Introduction to Deepseek, Qwen, and Gemini models (1h)
- Presentation of the three models: origins, strengths, weaknesses, use cases.
- Comparison with GPT-4, Mistral, Claude and other open-source models.
- Open-source AI vs. proprietary AI: what are the criteria for choosing a model?
2. Experimenting with models via OpenRouter and API (1h30)
- Introduction to OpenRouter: what is it and why should you use it?
- Compare model performance by making API requests.
- Practical case : sending the same request to different models and comparing the responses.
- Presentation of tools for testing API calls: Postman, Python (requests), OpenRouter Playground.
3. Confidentiality and data management (1h)
- What data are recorded by these models?
- Models hosted in the cloud vs local execution: what are the challenges?
- Encryption, anonymization, and data protection best practices.
- Use cases where to avoid SaaS AI and favor local solutions.
4. Automation with Make: when and how to integrate AI models? (1 hour)
- Introducing Make to automate AI workflows.
- Examples of automations:
Automatic generation of summaries via OpenRouter.
Sending AI responses in a CRM or shared document.
Connect Make with Notion, Google Docs, Slack. - Practical case : creation of an automated scenario with an AI model.
5. Running your own AIs locally: LM Studio and Ollama (1h30)
- Why run a model locally?
Maximum privacy.
Performance: avoid the latency of a remote API.
Zero cost after installation. - Presentation of the tools:
LM Studio: simple interface for running open-source models.
Ollama: simplified management and execution of AI models locally. - Practical case : test an open-source model locally and interface it with a Python script.
6. Final comparison and recommendations for use (1 hour)
- Summary of the performances and benefits of Deepseek, Qwen, Gemini.
- Which model for what need?
Text summary
Editorial assistance
Code generation
Translation and reformulation - Interactive discussion: what model should be adopted in business or for personal use?
7. Questions/Answers and assessment with the KISS method (30 min)
- What participants are going to keep (Keep), improve (Improve), improve (Improve), start (Start), and stop (Stop).
- Development of an individual action plan: what AI should be integrated into its workflow tomorrow?
.png)
.png)
.png)
.png)
.png)
