You are viewing limited content. For full access, please sign in.

Discussion

Discussion

How we're deploying custom AI solutions for self-hosted Laserfiche clients

Self-Hosted Integrations
Updated July 28
posted on July 15

We are currently deploying AI solutions for some of our on-prem Laserfiche clients. I'm putting together a video series on how we've done it when client requirements go beyond the on-prem AI options currently available — e.g. full control over AI prompts and workflow, custom tasks like document comparison or a client who already has an established AI provider they need to build on.

The video below introduces the architecture (REST API, n8n, AWS Bedrock) and demos a few of the capabilities live.

https://youtu.be/4FV9cWBf768

I plan to add videos as frequently as I can get to them covering the details in more depth. Questions welcome.

5 0
replied on July 28

Nice job! Can't wait to see the rest of the videos and really dig into the finer details of your setup. A lot of us devs have trouble giving good presentations, but I think you've done really well with your pacing and breakdown of the overall topics. It's stuff like this that enables on-prem setups to not only stay viable, but opens up new ground for power users.

0 0
replied on July 27

Here's the second video: AWS Bedrock Setup

https://youtu.be/KKebbQiyilo

This starts with creating an AWS IAM user for Bedrock (you will need to set up your AWS account first) and ends with implementing the Bedrock model into a simple chat interface.


It takes more than I had anticipated to push out a video as it isn't something I have a lot of experience with.  These may take longer than expected.  Hopefully they will be helpful to some.

1 0
replied on July 17

Excellent work! We've just begun experimenting with 3rd party AI in Workflow and right now all we are looking at is just identifying document types/assigning templates and extracting field values. The AI chat feature you've introduced for self-hosted fills a really big void though. Really excited to go through the whole series of videos as they come out.

0 0
replied on July 27

I hope it all goes well for you.  I'll be covering classification in a couple more videos.  One thing Laserfiche did that can really help is to provide template descriptions.  I think the limit in on-prem is 200 characters (as opposed to 1000 or 2000 in cloud) which is still enough to provide the AI prompt some classification help.  Before classification, I read all the possible templates and their descriptions to build a solid prompt for the classification step.  Fields also have descriptions that can help guide the prompt when extracting field data from text.  Using these has significantly improved accuracy for this part of the process.

1 0
You are not allowed to follow up in this post.

Sign in to reply to this post.

Details
Related Posts
Loading ...