Build your own Local 'Perplexity' with Ollama – Deep Dive

Local Llama 3 Custom Web Search Agent with Ollama – Deep Dive
Join me for another technical deep dive as I walk you through how I built a custom web search agent that runs entirely on your local machine.

Chapters
Introduction: 00:00
Agent Schema: 01:10
Python Code Walkthrough - Overview: 04:00
Python Code Walkthrough - Prompts: 05:00
Python Code Walkthrough - WebSearch Tool: 14:30
Python Code Walkthrough - Agent: 23:14
Setting up Ollama: 32:38
Testing with Llama3 8B: 34:30
Testing with Code Lllama: 42:10
Testing with GPT-3.5-Turbo:51:35
Thoughts on Open-Source models: 55:17

[Data-Centric]

Update to Script:
I discovered that performance could be slightly improved by prompting the Planning agent to explicitly generate search engine queries instead of generic questions! Changes have been merged to the GitHub repository.

[lgdlgre]

bro i love your content, you're on of the few people providing real info on this topic, what do we need for a full series of custom agents? !!!🙌🙌

[freedtmg]

you are quickly becoming my favorite creator in this space.

[Sarvesh_Ganesan]

Extraordinary. Just what I've been searching for. A lot of youtubers build agents from other frameworks. THIS IS GOLDEN.

[deathdefier]

You sir are a pillar of this community, thank you <3

[Techtantra-ai]

bro u r the first creator who work with not only GPT in the name of AI

[augustus]

Your content is one of the best I've ever seen on this topic.

[supercurioTube]

I'm still watching, enjoying your presentation and that you're showing a project simple enough to follow easily.
And ouch, that first result is painful hahah, but it's great to show that it's not exactly trivial to get agents to do what we actually want.

`ollama list` return the list of installed models by the way

[MichaelWoodrum]

I like the thorough conversation. Thanks.

[okich]

Zero bullshit content. Many, many thanks!!!

[free_thinker]

Pure quality content as usual ❤ keep up the good work man!

[SilentMajority]

Clear, concise and broken down in great detail. Thank you and can't wait to learn more.

[trsd]

Wow, that’s amazing! You coded an open source perplexity! How cool is that 🤩

[madhudson]

Great video. Langgraph would be perfect for this kind of flow

[wadejohnson]

John, you are like the adult in a room of kids. Thank you for these intelligent, well-considered videos. Stay true to yourself. BTW, I ran the code using GPT-4o and it returned the correct answer and a good citation. The generated queries were also good. So, I would guess that doing agentic workflows successfully will require using bigger models.

[NoCodeFilmmaker]

Once again, amazing content bro. Can't wait to see you get to your first 100K 💪

[juanpasalagua]

Thanks for all the great info you provide!

[malikrumi]

Looking forward to those future videos you mention at the end. I prefer open source whenever possible. p.s., you were using raw, not fine tuned models in these tests today. Perhaps you could also add another tutorial on domain specific, task specific, fine tuning of open source models.

[SolidBuildersInc]

Great content,

Very nice script. I follow the logic pretty well. How would you implement a RAG solution for private data?

It appears the structure workflow for the agent protocal could be used for a pdf, db, xls etc. So ex. A citation for a 22 page pdf would be like page 13. Paragraph 2 etc.

Please advise how this RAG could be done dynamically to minimize the complexity of the code, if you have thought about this.

Thanks for Sharing

[jakeparker]

I've had the best luck constructing the json and dictionary with classic code based on whether or not the LLM thinks the content of a specific page answers a question.