This AI Agent with RAG Manages MY LIFE
Vložit
- čas přidán 9. 09. 2024
- Artificial Intelligence is no doubt the future of not just software development but the whole world. And I'm on a mission to master it - focusing first on mastering AI Agents.
In this video I show how to implement an AI Agent with RAG that can search through your meeting notes and create tasks in your favorite task management software based on action items in the notes! Along with that, it does all the task management you would ever need. I personally use this myself to save a ton of time and help keep me organized!
All code for this RAG AI Agent can be found here:
github.com/col...
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
If you are curious about using other types of LangChain document loaders to chat with other types of documents, check out:
python.langcha...
If you are curious about using other Vector databases besides Chroma with LangChain, check out:
python.langcha...
Take this documentation and run with it! LangChain gives you everything you need to create super robust and production ready RAG applications.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Join me as I push the limits of what is possible with AI. I'll be uploading videos twice a week - Sundays and Wednesdays at 7:00 PM CDT!
Incredible! Seeing an AI agent manage daily life tasks is engaging. AI agents are truly changing the way we live and work.
I appreciate you skimming the environment setup and getting to the meat. Seems like most tutorials are 90% are “install this”, “pip install that” with 10% code explanations. Good work
@@j0hnc0nn0r-sec Haha I agree and that's what I'm shooting to avoid. Thank you!!
Quality Information. Relevant to my interests. Liked + Subscribed! Looking forward to the Cloud Tutorial!
Thank you for sharing knowledge!
Thank you very much Jared - I appreciate it a lot!
Great content! you deserve more subscriber!.
@@yupickmyusername Thank you - that means a lot!
Awesome work Cole. Just out of interest, what are your thoughts on Graph RAG and also semantic search vs similarity search? Maybe a combination of all three might be relevant in some use cases. It would be cool to have the Agentic workflow figure out the optimum approach 🙂, maybe it'll throw in a fine-tuning request too!
Thank you, Jonathan!! I wanted to keep things nice and simple for this guide, but I definitely have a lot more content coming on RAG in the future where I'll dive into things like graph RAG and semantic search to improve results. I absolutely love your idea of having an agent figure out the best kind of RAG to use based on a prompt!
"My AI ate my homework."
Awesome can you one for Todoist?
I'll definitely consider it! Is that something you use over Asana for task management?
Yeah, but can you use local llm, like ollama?
@@kyborq You sure can! I'll be making more videos soon using local models for RAG agents from Ollama and HuggingFace!
nice! just wondering re embedding, why not use triplex so that you can do global context-type searches?
Thanks and great question! I wanted to start with something super basic for the embeddings, but using Triplex is definitely a great idea! Honestly, I could make a whole video just on embeddings, and using specific tech like Triplex to improve results would be a huge part of that.
@@ColeMedin yes, why not make (a) vid(s) about embedding(s)? it's a critical part to get relevant info back. question is, is triplex the one-for-all, or should the data to be ingested be evaluated and then nomic used for those that don't have a global context, vs triplex for those that do, what about ms' graphrag, is it still relevant? also, is it possible to "stack" embedding models? imo making a "rag router" makes more sense, where "use nomic if not global context within the data to be ingested, otherwise use triplex"... what do you think? imo a vid w/ test would be awesome!
@@themax2go I absolutely love your thoughts here - thank you!
I agree embeddings are critical - I will definitely be making a video on RAG focusing on embeddings in the near future.
Your idea of a "RAG router" was actually suggested by someone else to me recently - I'm starting to see a trend emerge haha
And I love it! A video to test this out with Triplex/Graph RAG/Semantic search/etc. is definitely in order 😉
you are rigth, the IA agents it's the key ...
Are you using that eyes always at the camera AI? Your eyes never leave the camera dead on even when you close your eyes and then open them.
I'm not actually! I've just practiced keeping eye contact with the camera because I've found it looks strange if I glance away.
I have tried using the "eye contact" AI tech before, but it glitched a decent amount so it ended up looking pretty strange haha
Little life hack, I put a tiny pair of cartoon sticker eyes around my Webcam lens. Find it much easier to look at the camera as though it is the person when I do that.
@@calmhorizons That's a great idea, thanks for sharing!
would this work with notion or with local versions like obsidian ?
@@JuankM1050 Yes this could be modified to work with really any software including Notion or Obsidian! You would just have to swap out the Asana API calls for Notion API calls. The tools and their descriptions could be kept very similar.
what is your desk? can you give some info about the desk?
I wish I could give you a good answer, but this is a virtual background haha so I'm not sure where to get this desk! One thing I will say though is I HIGHLY recommend getting a standing desk. Mine is a Vivo standing desk which has worked perfect for me for over a year.
The desk image is AI of course but it looks like a fun design to custom-build from scratch!
bro you look like AI
@@datasciencetoday7127 haha lmk what makes you think that so I can change it 😂
They already invented this years ago, it's called a calendar, and guess what? It doesn't hallucinate! How great is that?
The calendar might not hallucinate, but Jim does when he books over another meeting on my calendar for the third freaking time in the same week 😂
i do hate capitalism too
i could not get the chromadb library to work with the latest python 3.12 something about wheels ? no idea :( atm im just loading stuff into the prompt , not very clever way of doing it ie: "from {list} give me the closest match to {input}" kind of thing, obviously limits me to llama's context window size too
Sorry to hear ChromaDB isn't working for you! Mind sharing the full error? Your method definitely works but you're right that you'll quickly hit a context limit (and it could create a "needle in the haystack" problem where the LLM fails to pick out the relevant information). I'd love to help you get Chroma working!
@@ColeMedin thats good of you , i think ive just got it working now, i needed this select the C++ build tools, ensure the latest versions of MSVC and the Windows SDK are checked.
@sausage4mash there’s a way to make sure that you can start fresh: look on CZcams for a video about installing and using Anaconda for your platform of choice. It’s not too hard to learn, and it lets you set up clean new environments for any development or scripting projects of your choice. Then you look up ChromDB and its preferred Python environment, and create a new conda environment with that version of python (also and make sure VSCode is using that environment when you kick off your script).