Rob Quiros PRO

caborabo's activity

• All
• Models
• Datasets
• Spaces
• Papers
• Collections
• Community
• Posts
• Upvotes
• Likes
• Articles

posted an update 5 days ago

view post

Post

184

Most enterprise AI is still guessing.

We boost attention on repeated tokens.
We connect chunks that look similar.
We extract “semantic” relationships from isolated sentences.

But we’re missing the point.

Business meaning doesn’t come from what data looks like. It comes from how it’s used.

🧩 A contract clause copied into a policy doesn’t mean the same thing.
📄 A word like “status” changes meaning from Jira to FDA filings.
📊 A graph without axes labelled is just noise.

Until enterprises can ground context window data in metadata, workflow, and policy, we’re just building smarter approximations.

Check out my latest blog here:
https://www.caber.com/blog/e74c56ee-a15b-443a-be7b-3f3b82c68406

See translation

• 

👀

1

1

+

Reply

reacted to their post with 👀 5 days ago

view post

Post

204

A little structure to your context window data can go a long way.

15% better answer quality ⚡ 35% ⚡ faster—including the time it took the tools to generate the graphs and schemas‼️

The pace of improvement in AI coding applications has been incredible, even so a little 💪🏻context engineering💪🏻 often matters more than the app or model.

📊 Real examples from a recent codebase cleanup we added:
🥇 Dependency graphs showing what calls what
🥈 Schemas for Pandas DataFrames

and saw better answer quality delivered ⚡ 35% ⚡ faster—including the time it took the tools to generate the graphs and schemas‼️

See the post for details on what we did: https://www.caber.com/blog/12744d4b-1d65-4e6c-886d-978c0f1317f5

🔍 This applies beyond code: documentation hierarchies, data lineage, system architectures. Data relationships provide context that unlocks reasoning.

See translation

• 

👀

2

2

+

Reply

reacted to their post with 👀 5 days ago

view post

Post

253

🚀 A PDF Parsing Method That Outperforms AI by 1000×

Parsing PDFs at enterprise scale is notoriously hard — metadata is missing, structures are inconsistent, and AI models often slow to a crawl when you throw millions of documents at them.

In a recent blog post, I shared how we can use data lineage to process data in PDFs 1000× faster than AI approaches, since PDFs are almost always the result of exporting from some other format.

One way to do this:
✅ Parse all non‑PDF documents first,
✅ Build a RAG (retrieval‑augmented generation) index from them,
✅ Perform simple (non‑AI) parsing on PDFs and cross‑check against the RAG,
✅ Only fall back to AI parsing if the data isn’t already known.

Another is to use a lineage-based tool like we developed at Caber.

👉 Read the full deep dive and try the code yourself here:
https://www.caber.com/blog/2e0a903d-caa2-4e93-a3a6-94636ec5ee2e

If you’d like to reproduce the results, the blog post includes runnable code and step‑by‑step instructions. I’d love to hear how you handle PDF parsing and document processing in your projects — feel free to share your approaches or improvements in the comments.

See translation

👀

2

2

+

Reply

posted an update 10 days ago

view post

Post

204

A little structure to your context window data can go a long way.

15% better answer quality ⚡ 35% ⚡ faster—including the time it took the tools to generate the graphs and schemas‼️

The pace of improvement in AI coding applications has been incredible, even so a little 💪🏻context engineering💪🏻 often matters more than the app or model.

📊 Real examples from a recent codebase cleanup we added:
🥇 Dependency graphs showing what calls what
🥈 Schemas for Pandas DataFrames

and saw better answer quality delivered ⚡ 35% ⚡ faster—including the time it took the tools to generate the graphs and schemas‼️

See the post for details on what we did: https://www.caber.com/blog/12744d4b-1d65-4e6c-886d-978c0f1317f5

🔍 This applies beyond code: documentation hierarchies, data lineage, system architectures. Data relationships provide context that unlocks reasoning.

See translation

• 

👀

2

2

+

Reply

posted an update 17 days ago

view post

Post

253

🚀 A PDF Parsing Method That Outperforms AI by 1000×

Parsing PDFs at enterprise scale is notoriously hard — metadata is missing, structures are inconsistent, and AI models often slow to a crawl when you throw millions of documents at them.

In a recent blog post, I shared how we can use data lineage to process data in PDFs 1000× faster than AI approaches, since PDFs are almost always the result of exporting from some other format.

One way to do this:
✅ Parse all non‑PDF documents first,
✅ Build a RAG (retrieval‑augmented generation) index from them,
✅ Perform simple (non‑AI) parsing on PDFs and cross‑check against the RAG,
✅ Only fall back to AI parsing if the data isn’t already known.

Another is to use a lineage-based tool like we developed at Caber.

👉 Read the full deep dive and try the code yourself here:
https://www.caber.com/blog/2e0a903d-caa2-4e93-a3a6-94636ec5ee2e

If you’d like to reproduce the results, the blog post includes runnable code and step‑by‑step instructions. I’d love to hear how you handle PDF parsing and document processing in your projects — feel free to share your approaches or improvements in the comments.

See translation

👀

2

2

+

Reply