I am new to LLM’s and want to do some projects for build my portfolio. Any suggestions about it? Currently found making a chatbot or something else with LangChain.
Hi @emre570,
I think a cool project to start is building a semantic search with an embedding model like the sbert models.
Then you could extend the project with a QA transfomer that uses span tagging to tag the answers in the texts found from you semantic search.
Finally, you can extend this with a larger LLM such as Zephyr, Mistral, Llama or similar to generate the long from answers (RAG - Retrieval Augmented Generation).
Here are some cool youtube channels that might inspire you further:
James Briggs, Code you own ai and Greg Kamradt
3 Likes
Thank you for your answer sir, one more question; Which playlists should I follow from these channels? I looked at them, found some playlists but I’m not sure which one to start. I found these two:
No problem! I think I would start here for the semantic search. It’s worth mentioning that James often works with Pinecone as a vector store, as he works for them. I haven’t used Pincone myself yet, chromadb and faiss are good alternatives. Huggingface datasets even supports indexing a dataset with faiss. (datasets with faiss)
I would not try to pretrain a BERT model at the beginning, but rather to fine tune it. At the very beginning, it is also enough to take the models as they are and see how they interact with each other in your project.
I think that once you’ve got the hang of it, you’ll quickly understand the other topics without any big problems and can freely look at which projects appeal to you.
2 Likes
Thank you for your help sir. I think these channels will enough for me for now 
1 Like
I like wikich.at, I think it is a nice project.
Source are public, here: GitHub - datastax/wikichat
It is based on LangChain, Astra db, and OpenAI. But you can adapt to any LLM easily.
2 Likes
You can also try this beginner-friendly project: LLM Project to Build and Fine Tune a Large Language Model.
1 Like