Hello, I am trying to connect my react app to a pipeline I have built(it takes in a pdf and adds it to a media set, then extracts text, and creates a new object in my ontology). Is this possible? Is it better to do everything within the AIP instead of building a separate frontend? I ideally just want to have the user upload their resume, type in a request, and output a list of results.
Hi there!
I don’t have too much knowledge on uploading files and triggering Actions via ontology-sdk. However, it seems like you want to run a pipeline on-demand from the ontology-sdk?
If so, I don’t believe this is currently supported. Writing a function to perform this operation outside of Pipeline Builder might be the way to go.
Yes, I essentially wanted to trigger the pipeline when a user uploaded their pdf.
Gotcha! What would that function look like, I still want to take advantage of the transform and “use_llm” nodes provided by pipeline
Hey! What kind of request are you expecting your user to input?
One thing you could do (but this would only work for uploading PDFs without having a custom request) is have a media set where users can upload their PDF. Use that media set in Pipeline Builder where you can use the use LLM node (and turn “skip recomputing rows” on). Then, on the output of the pipeline, put a schedule on it so that it runs whenever the input dataset updates. I left out some details here but happy to talk more about this approach if it works for you.
I think you could also do something with Workshop (there’s an upload widget) and potentially use AIP Logic to look at media there but cc-ing the Media/Unstructured team on more advice cc: @Isy
Hi, ok so at a high level what I have envisioned is that either through some custom frontend or directly within AIP(whichever works best), a user uploads a pdf of their resume. The text is extracted and filtered into useful categories(School, major, skills, location, etc).
Then I have some Ai agent look at the filtered info(right now im thinking of making it a Resume Object in Ontology). The agent will be supplied with some context like “The user wants to find insider connections relating to Internships at Palantir” and choose a set of endpoints to call from linkedIN API.
Then somehow the results will get dumped into Ontology and another ai agent will perform semantic search or ranking to get the most useful search results from the LinkedIN API.
Im open to using a pipeline, for the initial PDF extraction process. Any suggestions on Tools I can use for the rest of the workflow would also be appreciated!
You can create an action allows users to upload media (to a media reference property) and you can then use OSDK to apply that action so that it is triggered via your custom react application. The release note for media upload being supported in OSDK is here and has useful code examples and documentation: https://www.palantir.com/docs/foundry/announcements/2025-02#media-is-now-supported-in-typescript-osdk-and-platform-apis-beta
When media gets uploaded, it gets written to your media set source. As Helen suggested, you could input the media set to Pipeline Builder and extract text in that pipeline. You can set up a schedule so that the pipeline runs whenever the media set input updates (ie. whenever media is uploaded). You could use the UseLLM board to process your extracted text or help filter into the categories you mentioned. Then your output dataset can back your ontology object.
Alternatively, you can use AIP Logic to extract text from media (the input would be an object with a media reference property), so you could avoid having a pipeline altogether and instead do all your processing there. You can run actions from within Logic so you could write back any information like the extracted text to your object.
In terms of providing an agent with information from the user’s request, I’m not 100% sure on the best way to do that, but someone else should be able to help there!
Thank you, I was able to set up the OSDK to automatically upload a pdf to the media set through my frontend. The media set triggers my pipeline through a schedule and outputs the object to my Ontology.
1 Like
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.