Activity
Mon
Wed
Fri
Sun
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
What is this?
Less
More
Memberships
AI Automation Society
208.5k members • Free
2 contributions to AI Automation Society
Jan 4 •
Enabling an assistant to process and discuss uploaded documents
I’m trying to reproduce a use case where the assistant processes a PDF file uploaded by the user. The file could potentially be very large. Once uploaded, the user should be able to ask questions about its content, similar to how ChatGPT operates. The file must not need to be permanently stored ... it will only be retained temporarily within the conversational context. What’s the best approach to achieve this in n8n? My idea is to break the file into smaller chunks and store them in a temporary buffer memory, such as a window buffer memory, allowing the agent to perform RAG searches on this temporary memory. Is this approach feasible in n8n, or is there a better alternative? How would you handle this scenario?
0 likes • Jan 4
@James Walls Thanks for the answer! I'm considering using a vector database, as you suggested, but the deletion part is what's tripping me up. What should be the trigger to delete the data? I'm considering using RediSearch with an expiration time, based on my research, but I'm also open to other approaches if there's a better one
0 likes • Jan 4
I solved it using Supabase, but I'm not entirely satisfied with temporarily storing uploaded files and deleting them after a few hours. Also, storing all the uploaded files together might confuse the RAG search, especially with large files. If a file has too many chunks, I can't ask an agent with top K = 4 to summarize it effectively (it wont read the whole document again). I'm considering storing just the summary of parts... I'll think it over
Dec '24 •
$fromAI Limitations with 'Call n8n Workflow' and 'HTTP Request' Tools
According to the documentation, it appears that $fromAI does not function with the "Call n8n Workflow" or "HTTP Request" tools, which is quite disappointing. Has anyone found a reliable workaround for this issue? I'd appreciate any suggestions!
0 likes • Dec '24
I intend to invoke the workflow tool using parameters systematically collected during the conversation with the agent. Throughout the interaction, the agent will engage with the user, gathering various inputs. Once all necessary information has been comprehensively collected, the agent will proceed to call the workflow tool, passing the input parameters individually—such as 'users_company', 'selected_product', and other relevant details.
0 likes • Dec '24
Thanks for sharing! I managed to solve it using prompting, but I'm not entirely confident in its reliability. While it worked in most tests, there were instances where the AI didn't follow the prompt exactly, causing it to fail. A more structured solution would likely be more dependable.
1-2 of 2
Active 296d ago
Joined Dec 9, 2024
Powered by