Hello. Can any one guide me in the direction of AI...
# random
n
Hello. Can any one guide me in the direction of AI-based tool development? For example, creating a chatbot based on a PDF. This is not a commercial project. Just looking to explore the AI.
n
There's a million and one articles on how to do this but if you are a dev and like to tinker this got me started: https://github.com/mayooear/gpt4-pdf-chatbot-langchain Not an article but an example of setting up and running your own chatbot based on a selection of PDF documents. Just getting this working on my machine , was, for me, a good experience and forced me to look in to the surrounding tech including understanding vector stores and prompt engineering.
n
Thanks @NElliott. I will definitely look into it.
n
There's also a tonne of free courses here: https://learn.deeplearning.ai/
👍 1
I have a raft of bookmarks from last year, too many to list out 😄
A lot of adverts on this site but it's quite clear and if you're comfortable with Python (which I wasn't / am not) then it'll be a breeze for you. https://beebom.com/how-train-ai-chatbot-custom-knowledge-base-chatgpt-api/ I managed to get it working.
👍 1
n
I will explore this one too. Can I reach out to you for help if needed?
n
Sure you can. I'm no expert and frankly not written any AI related code in months but happy to help where I can. (also note I'm in Devon from Saturday for a week on holiday 😉 )
😄 1
j
@NElliott - Was it you that integrated AI into netsuite for financial transactions? Sent them off for analysis and returned a summary?
n
@Jono I used AI to analyse the results of a saved search, yes. I had a simple search with a sales reps totals and their names. It was able to understand what the data represented and I had it present to me how this data might be useful to a CEO / CTO / Marketing person etc. (bullet pointed list of key takeaways) I then extended this so a user could choose a saved search. It would give an initial overview of the data and then I allowed the user to add questions. A big limitation is the size of the data since I was passing this in with each request so that it could be analysed. This is not very workable but gave me an insight in to how it could be built going forward. Using "agents" might be a better way and also I feel sure there was a neat way to handle the token limit but I've not touched that piece of code in months.
j
interesting; Does it get used heavily by the finance team? I've always thought it would be good for the finance team to be able to have AI give a why on a P&L variance report. (ie. xx transaction for $ was unusual and accounted for x% of the variance)
n
No, it doesn't get used at all it was just an experiment by me to see what I could do with AI inside NetSuite. 🙂 There's a token limitation when you make calls to the LLM, you can think of it as a data limit, and most searches beyond something basic, just output as raw data, get pretty big quickly. I never had the time to explore token limitations otherwise I might have taken it further. The chatbot I developed, based purely on NetSuite documentation / SuiteAnswers on the other hand, is used by the business daily and even some of our customers.
It looked like the screenshot. When you selected a saved search it'd pass the data to OpenAI and ask it to describe the purpose of the saved search and a brief overview of the data. You could then ask questions of the data. I was just trying it now to give you an example of it in action but forgot that it's calling OpenAI and we moved to using Claude a while ago. 🙂
j
very neat; a chatbot is on my ideas to explore - would be nice to take a volume of queries away from a helpdesk