ArnavAgrawal03

5 ms·

1.
by ArnavAgrawal03 10mo ago
you can do that with Morphik already :) We use an embedding model that processes videos and allows you to perform RAG on them.
2.
by ArnavAgrawal03 11mo ago
Came here to say I love Ocaml too
3.
by ArnavAgrawal03 1y ago
We use a BSL for our product ( https://morphik.ai ) and usually stay away from calling it anything. We'd just say "repo is public at: https://github.com/morphik-org/morphik-core ". I like the te
4.
by ArnavAgrawal03 1y ago
we used multi-vector models at Morphik, and I can confirm the real-world effectiveness, especially when compared with dense-vector retrieval.
5.
by ArnavAgrawal03 1y ago
> They had known him for only 15 seconds, yet they still perceived the act of snapping him in half as violent. This is right out of Community
6.
by ArnavAgrawal03 1y ago
Our argument in general is that even in the non-flattened cases, we see complex diagrams pop up in documents that won't work with a text-based approach. In the context of RAG, the objective is to send information to the model, so LLMs
7.
by ArnavAgrawal03 1y ago
Would love to try our hand at it! We have a couple magazine use cases, but the harder it is, the more fun it is :)
8.
by ArnavAgrawal03 1y ago
Would love feedback :)
9.
by ArnavAgrawal03 1y ago
Yes! We have a use case in production with over a million pages. MUVERA is good for this, since it is basically akin to regular vector search + re-ranking. In our current setup, we have the multivectors stored as .npy in S3 Express storage.
10.
by ArnavAgrawal03 1y ago
For HTML, in a lot of cases, using the tags to chunk things better works. However, I've found that when I'm trying to design a page, showing models the actual image of the page leads to way better debugging than just sending the c
11.
by ArnavAgrawal03 1y ago
Completely agree with this. This is what we've observed in production too. Embedding images makes the RAG a lot more robust to the "inner workings" of a document.
12.
by ArnavAgrawal03 1y ago
You can add OCR with Gemini, and presumably that would lead to better results than the OCR model we compared against. However, it's important to note that then you're guaranteeing that the entire corpus of documents you're pr
13.
by ArnavAgrawal03 1y ago
Multimodal RAG is exactly what we argue for. In their original state, though, multivectors (that form the basis for multi-modal RAG) are very unwieldy - computing the similarity scores is very expensive and so scaling them up in this state
14.
by ArnavAgrawal03 1y ago
This would depend on the exact use case. Feeding in the invoice directly to the model is - in my opinion - the best way to approach this. If you need to search over them, then directly embedding them as images is definitely a strong approac
15.
by ArnavAgrawal03 1y ago
That's an interesting point. We've found that for most use cases, over 5 pages of context is overkill. Having a small LLM conversion layer on top of images also ends up working pretty well (i.e. instead of direct OCR, passing batc
16.
by ArnavAgrawal03 1y ago
I've used your product and particularly like that you show bugs in a table instead of littering my entire PR. Does Jazzberry run on the entire codebase, or does it look at the specific PR? Would also like some more details about the to
17.
by ArnavAgrawal03 1y ago
While we don't use this yet, it seems very promising - thanks! We did something similar (with libreoffice, for example) to have support for non PDF datatypes, but this seems like it is coming at it more from a security perspective - wh
18.
by ArnavAgrawal03 1y ago
totally agree that air-gapped provides unparalleled peace of mind. That's a major reason why we have strong support for local deployment. Nice to know that our hypothesis is somewhat accurate :)
19.
by ArnavAgrawal03 1y ago
No, you can bring your own LLM. In the cloud, we're querying gpt-4o. We're looking to expand to have some fine-tuned VLMs for document parsing and extraction further in the roadmap, but that would heavily depend on use-case.
20.
by ArnavAgrawal03 1y ago
Yes! If you're running the local version and it's taking long, that an indication that your GPU isn't being used properly. This can be traced back to the `colpali_embedding_model.py` file, where you can set the device and att
21.
by ArnavAgrawal03 1y ago
Hey! what format of files are you uploading? seems to work ok on my end...
22.
by ArnavAgrawal03 1y ago
Thanks for the feedback :) We're using the MCP daily as we develop Morphik further. Thought it would be a nice thing to share. "I used the stones to [make] the stones" haha
23.
by ArnavAgrawal03 1y ago
Thank you!! It works particularly well with those. We use ColPali-style embeddings for our visual doc search. As a result, we're not limited by parsing quality the same way typical RAG systems are. Here's a link to a blog I wrote
24.
by ArnavAgrawal03 1y ago
Just adding some of our roadmap here as well: - AST parsing and conversion to visual graphs for easier understanding of large codebases - Integrate custom knowledge graph editing, parsing, and retrieval with Morphik MCP - Slack, Jira, and C
25.
Ask HN: What RAG evaluations do you care about?
1 points by ArnavAgrawal03 1y ago|0 comments
26.
by ArnavAgrawal03 2y ago
that's some really good feedback! Will definitely add a "Deploy your first app with Morphik" section. You can definitely use this with products like obsidian. You might find out graph offering particularly helpful in that cas
27.
by ArnavAgrawal03 2y ago
Try it out: - Clone the repo at: https://github.com/morphik-org/morphik-core - Launch the UI component following instructions here: https://docs.morphik.ai/using-morphik/morphik-ui
28.
Show HN: I built an open-source NotebookLM alternative using Morphik(github.com)
23 points by ArnavAgrawal03 2y ago|3 comments
29.
I gave GPT 4.5 my journal to test its emotional intelligence(twitter.com)
3 points by ArnavAgrawal03 2y ago|0 comments
30.
Show HN: I built a vision-native RAG pipeline(databridge.mintlify.app)
2 points by ArnavAgrawal03 2y ago|1 comments

More ›