Créer un Enseignant AI : Mon Voyage avec les Bases de Données Vectorielles et ChatGPT
Bien sûr, je peux vous aider. Cependant, vous n’avez pas fourni la transcription de la vidéo. Veuillez coller le contenu ici et je serai heureux de créer un résumé pour vous.
Source : Gaurav Sen | Date : 2024-11-30 04:30:21 | Durée : 00:13:43
Have been watching your content for 4 years now and in every new video I learn something new. Thanks a lot Gaurav for working hard for us. I genuinely want to thank you for continuously sharing your content for free on YT.
Awesome video! Thanks Gaurav, you just got a new subscriber! 🙂
I have read all the comments till 9th of jan 2025
Good afternoon sir,
I am student at online degree of IIT Madras i.e. BS in data science and applications,
I want to make a teaching assistant like tool for them,
But students may use it to solve their assignments also, though they are doing the same thing now,
but teachers and staff of iit madras would never approve this officially,
so what can be solution to this problem from your side,
we are also provided with transcripts of lectures by the team of iit madras, i can use them.
and they are also offering 5 dollars api credits per months for 1 year,
so I can use them for learning purpose by implementing this project.
Thank You sir, I got this idea by watching your this video.
I am in first year of my college,
But i will do it anyhow, not just doing it but learning all the tech that are coming in my way.
thnq
This is a good and simple approach Gourav. But a even simpler approach could have been – using MLT feature of ES for fetching the recommended videos related to the query
It is opening doors of my mind and teaching assistants alternative
What an idea for byjus and edtechs
When you are attaching the entire file, it would be large right, there can be the content in multiple videos? Doesnt it cost more tokens? There can be videos with large content, attaching the entire transcript would cost more, because you are just identifying the relevant transcripts not the chunks in transcript which are relevant right
This is great. I used qdrant db for similar things. Have a look at that also and let us know your analysis
Nice Video Gaurav, would love to see a hands on demo.
Suppose I am looking for using this system you described to build sql queries, in such case I need to upload the DDL of the tables as files in neon and same DDLs in ChatGPT as files. In such case can I get query from gpt generated with prompt I give in natural language for eg. Give me top 10 customers. Will this be giving me correct query.
I think AWS Kendra does the same. It’s not cheap though.
how to make it history context aware multi user bot?kindly cover that
Great one. Keep it up.
Very insightful
I somehow feel vector database here is an overkill.. why cant a recommendation search work here with content filtering? may be you could have directly used elastic search for getting relevant recommendations for the topic and then pass it on to Open AI to get appropriate answers… or even graph database would suffice
Good content ❤
Is this same as customgpt knowledge base file ? I mean when would you use customgpt knowledge base and vectors?
I think you can do it with Notebook LLM easily
Do you have GitHub for this ??
Why you have not used Amazon Aurora for interview-ready database?
How are the context files generated? I understand that the transcription is done by AWS but for a given query how are files with relaxant context chosen? In other words what's the input to Neon DB? Please excuse me if my question doesn't make much sense totally new here. Keep up the good work 👏🏾
A variation of this approach would be to pass the response from ChatGPT to a fine-tuned text to speech(TTS) model(say AWS Polly) to mimic Gaurav's voice, intonation and delivery style. Learner's come for Gaurav's teaching style and this would greatly enhance the user experience.
One other extension would be to make this real time. Where the learners can pause the lecture at any point and ask a question contextually at that point. This would mimic the experience of a live class where the student can get his query solved interactively.
Awesome content. I have been searching for this content for the past one month and just found it 😂. And it is only 13 minutes?. Straight to the point. Thank you. God Bless.
Great Video. But some finer details or a deep-dive part-2 might be helpful. In your demo, you manually added a system prompt with the files. But in a real chat scenario you would generate the system prompt after querying the Neon DB. Is that left as an exercise to the viewer?
Great, thanks