Crawl4AI and Instructor in a 12-tool AI stack overview

This Persian overview introduces 12 tools for building AI applications, with an emphasis on RAG and running software on your own system. The speaker presents them as open source alternatives to paid services, but does not walk through installation or benchmark their performance.

Ollama is the suggested starting point to run models locally. The speaker describes its command-line workflow and OpenAI-compatible API, then contrasts personal use with vLLM's focus on serving many concurrent users and improving GPU utilization.

The document pipeline starts with Chonkie for splitting text into chunks. Marker converts documents into Markdown; the speaker says it preserves document structure and supports a local document-processing route. Qdrant stores vectors and supports searches using meaning and metadata. Crawl4AI handles browser-based web extraction, including JavaScript pages and interactions such as scrolling and clicking.

Other sections cover DSPy for prompt optimization, Outlines for constrained output, and Instructor for schema validation and retries. LiteLLM provides a common API across model providers. Langfuse records requests, responses, timing and costs, while Ragas supplies metrics for evaluating retrieval and answers.

The closing caveat matters: the speaker says they have used only some of these tools. The remaining descriptions are introductions rather than reports of firsthand testing.