Adityav369

5 ms·

1.
by Adityav369 1y ago
You can ask the model to describe the image, but that is inherently lossy. What if it is a chart and the model gets most x, y pairs, but the user asks about a missing "x" or "y" value. Presenting the image at inference i
2.
by Adityav369 1y ago
Yeah the fine tuning is definitely the best part. Often, the blocker becomes high quality eval sets (which I guess always is the blocker).
3.
Don't bother parsing: Just use images for RAG(morphik.ai)
328 points by Adityav369 1y ago|72 comments
4.
by Adityav369 1y ago
Mainly to maintain layout information. Also search becomes easier this way.
5.
by Adityav369 1y ago
We do use ColQwen! Currently 2, but upgrading to 2.5 soon :)
6.
by Adityav369 1y ago
Yeah we had an overload on the ingestion queue. If you try again will be much faster as we just moved to a beefier machine. (The previous ingestion will still work since it is in queue, but new ones will be faster)
7.
by Adityav369 1y ago
Depending on the use case, it happily runs on my MacBook air M2 16GB ram with mps for small pdfs, and searching over 100-150 documents with colpali takes a 2-ish minutes. Very rough numbers. For ingestion, takes around 15-20-ish seconds a p
8.
by Adityav369 1y ago
For ingesting graphs, you can define a filter, or certain document ids. When updating, we look at if any other docs are added with that filer (or you can specify new doc ids). We then do entity and relationship extraction again, and do enti
9.
by Adityav369 1y ago
You can run this fully locally using Ollama for inference, although you'll need larger models and a beefy machine for great results. On my end llama 3.2 8B does a good job on technical docs, but bigger the better lol.
10.
by Adityav369 1y ago
Thanks we should have been more clear. The part in ee is our UI, which can be used to test or in dev environments. The main code, including API, SDK, and the entire backend logic is MIT expat.
11.
by Adityav369 1y ago
Do you mean ingesting the extracted rectangles/ bounding boxes? We're actually working on bounding boxes, this is a good insight and we can add this to the product. However, the way we ingest is literally converting each page to a
12.
by Adityav369 1y ago
Depends on your document types. If you're using txts, then plain RAG built on top of any vector database can suffice depending on your queries (if they directly reference the text, or can be made to, then similarity search is good enou
13.
by Adityav369 1y ago
We have two ingestion pathways: 1. regular OCR + text embeddings; 2. Colpali. We've observed that Colpali does a much better job with tables since it can encode positional stuff and layouts as well.
14.
by Adityav369 1y ago
The paid part applies to the ui-component which provides a chat user interface. The core code, SDK, api is all under MIT license.
15.
by Adityav369 1y ago
Thanks for pointing that out! Fixed it. For the metadata extraction, we save these as Column(JSONB) for each documents which allows it to be changed on the fly. Although, I keep wondering if it would have been better to use something like m
16.
Show HN: Morphik – Open-source RAG that understands PDF images, runs locally(github.com)
200 points by Adityav369 1y ago|43 comments
17.
by Adityav369 1y ago
Apologies for the poor formatting! Features: - Multimodal search across text, diagrams, and videos - Natural language knowledge base management - Fully open-source with responsive support
18.
Show HN: Morphik – Open-source MCP server for technical document search(docs.morphik.ai)
7 points by Adityav369 1y ago|6 comments
19.
Show HN: I built an open-source vision RAG that embeds images and text together(docs.morphik.ai)
4 points by Adityav369 1y ago|0 comments
20.
Show HN: DataBridge: Open-Source Multi-Modal Parser and Retriever for LLMs(github.com)
2 points by Adityav369 2y ago|0 comments
21.
Show HN: DataBridge: Rule-Based Metadata Extraction, PII Redaction, and More(databridge.gitbook.io)
4 points by Adityav369 2y ago|0 comments
22.
Show HN: Build video AI agents with DataBridge(youtube.com)
8 points by Adityav369 2y ago|0 comments
23.
by Adityav369 2y ago
Good question! You can use any open source or private model you want with this (just by changing one line in `databridge.toml`), whereas notebook lm is limited to gemini. While our UI component allows it to be used as an open source noteboo
24.
Show HN: I built an offline open-source RAG system DataBridge(github.com)
7 points by Adityav369 2y ago|2 comments
25.
Show HN: DataBridge - An open-source, modular, multi-modal RAG solution(github.com)
5 points by Adityav369 2y ago|0 comments
26.
by Adityav369 2y ago
Thank you. Definitely! We have cross posting for linkedin and instagram as of now. Integrating ActivityPub/ Fediverse is on our list. Have you personally faced this issue/ know of anyone who has been looking for a solution to fill
27.
by Adityav369 2y ago
How to use/demo video: https://www.youtube.com/watch?v=Q6twwYsYNKI&t=1s Hey HN! My brother and I are building BrandSync AI to address a problem we identified while talking with businesses: Businesses have existing
28.
Show HN: Automate your social media workflow(brandsync.ai)
3 points by Adityav369 2y ago|4 comments