vault backup: 2025-07-21 14:56:41

Affected files:
.obsidian/plugins/text-extractor/cache/0e1681cb976efab437314ee5b955bf19.json
.obsidian/plugins/text-extractor/cache/8e7ba73749fd1e85200c90d920c18377.json
.obsidian/plugins/text-extractor/cache/e3491f3656f5e7a8c562d164a907bfc6.json
.obsidian/workspace.json
0 Journal/0 Daily/2025-07-21.md
Attachments/Pasted image 20250721140924.png
Attachments/Pasted image 20250721140942.png
Attachments/Pasted image 20250721142129.png
Temporary/AI Agents.md
This commit is contained in:
2025-07-21 14:56:41 +02:00
parent 071b90b50f
commit 9c67d2fdec
9 changed files with 79 additions and 11 deletions

51
Temporary/AI Agents.md Normal file
View File

@@ -0,0 +1,51 @@
---
title: AI Agents
created_date: 2025-07-21
updated_date: 2025-07-21
aliases:
tags:
---
# AI Agents
## [He created an AI to do his job (his boss is hallucinating) - YouTube](https://youtu.be/MOlO1_mj1dU)
### Agents
![[Pasted image 20250721140924.png]]
![[Pasted image 20250721140942.png]]
- anything-llm
- crew-ai
- fine tuning vs RAG (RAG is metter?)
- langchain
#### Agent Roles
Google Ads Optimizer
- research assistant --> search the good keywords
- content strategist --> B2B or B2C
- manager: prompted.
![[Pasted image 20250721142129.png]]
- double checks output of the two agents before sending it to the google API. If there is a formatting error in the csv file or so it just asks the agents to restart.
##### React Example
- input figma (wireframe prototype): really nice application overview with all screens etc
- asked vision model to describe the different screens in detail (Pixtral 12B)
- Get Prompts (1-20) --> one super prompt file with all prompts inthere
- send to agents which execute stuff on cursor
- you need the agents because of the context length limit
Roles
- prompt manager: prompt analysis --> split prompt into multiple chunks which match best with the agent team and send them the prompt with a global context description
##### Cursor
- Combine with crewAI
### Bikee Example
- Reverse Engineering crew --> creates a document which describes
- brute force routes
- Developer crew
### Learnings
- use AI to make prompts
- combine