amrutha srivatsav

research @databricks

[amrutha srivatsav @amruthaaa___]

can’t wait to run it back!! 🎆

[ayushe @ayushe_n]

hosting diwali in sf again🪔 last year 150 people came. looking for a venue for this year — anyone have a space or know someone who does? dm me :)

[amrutha srivatsav @amruthaaa___]

@ayushe_n BYEEE

[amrutha srivatsav @amruthaaa___]

@tomaarsen @DbrxMosaicAI The endpoint we used is listed here! https://t.co/OLzQDnkKMi (databricks-bge-large-en)

[amrutha srivatsav @amruthaaa___]

Charts contain important information in enterprise documents, but agents often struggle to retrieve and reason over them accurately. Chart images can be misread, while captions often omit the precise values and labels needed for retrieval.

That’s why our team at @DbrxMosaicAI developed structured chart extraction for ai_parse_document, converting chart content into JSON that can be indexed and used by Genie and other RAG systems.

Across two chart-heavy benchmarks, our approach, which uses a lightweight 300M-parameter text embedding model, outperformed multimodal embedding baselines which were up to 10× larger.

[amrutha srivatsav @amruthaaa___]

@aaxsh18 @DbrxMosaicAI Hi @aaxsh18 ! We are not planning to release the dataset, as we created it with the intent of testing more realistic scenarios at a smaller scale rather than as a full benchmark. However, the documents we created the questions from are linked in the blog post!

[Michael Bendersky @bemikelive]

Cool counterintuitive result from our team: on 2 hard chart-retrieval benchmarks, a small text embedding + up-front parsing (ai_parse_document) matches or beats a big multimodal embedding model.

Big congrats to @amruthaaa___ , who led this work! 👏
https://t.co/Bl0W6RCH5H

[amrutha srivatsav @amruthaaa___]

@vibhayellamraju 😍😍

[amrutha srivatsav @amruthaaa___]

@DbrxMosaicAI Read more about the work here! https://t.co/9tLncApQyD

[amrutha srivatsav @amruthaaa___]

@vibhayellamraju now they r just divas tbh

[amrutha srivatsav @amruthaaa___]

@vibhayellamraju weekend of good food

[amrutha srivatsav @amruthaaa___]

@vibhayellamraju yes

[amrutha srivatsav @amruthaaa___]

@vibhayellamraju LOVEEE

[amrutha srivatsav @amruthaaa___]

@vibhayellamraju best coffee

[Jane Zhang @jjanezhang]

Are you using Claude Code or Codex to:

- Ask questions over a dataset of PDFs
- Turn your docs folder into structured tables or JSON
- Classify documents like customer notes over a set of labels

If so, you could wait an hour for Claude to run a for loop + multimodal model on your documents

or you can add the Databricks ai_parse_document function to one of your skills and batch process your documents faster and more accurately 💃🏻

[Krista Opsahl-Ong @kristahopsalong]

Awesome opportunity for students excited to apply their research to real-world enterprise problems — and compete for $100k in model credits 💰

Apply by April 3 → https://t.co/qBmWGVmqTg