Hacker News: prats226

New comment by prats226 in "Research-Driven Agents: What Happens When Your Agent Reads Before It Codes"

prats226 — Thu, 09 Apr 2026 21:09:59 +0000

A good experiment would be to also try giving it access to latency traces so it can identify issues? Wrt coding agents, giving access to observability tools often improve coding/debugging ability for me

New comment by prats226 in "Ask HN: Alternatives to Reducto?"

prats226 — Tue, 03 Mar 2026 07:04:01 +0000

Try https://docstrange.nanonets.com/ once, 10k docs you can use for free. Strong table performance. Do give feedback if any. Powered by bigger model compared to our open source one which is quiet popular on HF.

New comment by prats226 in "Large-Scale Online Deanonymization with LLMs"

prats226 — Wed, 25 Feb 2026 22:56:10 +0000

If with LLM's you can deanonymize at scale, on a personal level, you should also be able to figure out what posts are leading to this deanonymization and remove them or modify them.

New comment by prats226 in "Show HN: Ocrbase – pdf → .md/.json document OCR and structured extraction API"

prats226 — Wed, 21 Jan 2026 00:48:15 +0000

Instead of markdown -> LLM to get JSON, you can just train a slightly bigger model which you can constrain decode to give JSON rightaway. https://huggingface.co/nanonets/Nanonets-OCR2-3B

We recently published a cookbook for constrained decoding here: https://nanonets.com/cookbooks/structured-llm-outputs/

New comment by prats226 in "LLM Structured Outputs Handbook"

prats226 — Sat, 17 Jan 2026 00:07:49 +0000

https://nanonets.com/cookbooks/structured-llm-outputs/uncons...

New comment by prats226 in "LLM Structured Outputs Handbook"

prats226 — Fri, 16 Jan 2026 23:49:48 +0000

Nice, it would be good idea to develop CFG for this as well so can embed it into all these constrained decoding libraries

New comment by prats226 in "LLM Structured Outputs Handbook"

prats226 — Fri, 16 Jan 2026 22:38:34 +0000

One of the authors here, will checkout the diagram link.

Every commercial model provider is adding structured outputs so will keep updating the guide.

New comment by prats226 in "DeepSeek OCR"

prats226 — Mon, 20 Oct 2025 20:09:37 +0000

https://docstrange.nanonets.com/ as well, wrapper on top of 7B version of https://huggingface.co/nanonets/Nanonets-OCR2-3B

New comment by prats226 in "DeepSeek OCR"

prats226 — Mon, 20 Oct 2025 20:08:02 +0000

Then you can just download finetuned version of same multi-modal foundation model that's trained on documents?

New comment by prats226 in "DeepSeek OCR"

prats226 — Mon, 20 Oct 2025 20:05:40 +0000

Top 3 models on huggingface are all OCR models. Most automation projects involve documents where you need a model finetuned to understand all elements inside documents and provide grounding and confidence scores etc which is why these subset of models are gaining popularity

New comment by prats226 in "What Americans die from vs. what the news reports on"

prats226 — Tue, 14 Oct 2025 23:42:10 +0000

Would be intersting to see where funding goes to fix these issues. News would heavily impact public opinion and hence political influence and public funding.

New comment by prats226 in "Nanonets-OCR2-3B – OCR model that transforms documents into structured markdown"

prats226 — Tue, 14 Oct 2025 23:34:57 +0000

Yes, and its not just OCR (Optical Character Recognition), it understands layouts, captures signatures, charts, watermarks etc so way beyond just characters

New comment by prats226 in "Launch HN: Extend (YC W23) – Turn your messiest documents into data"

prats226 — Sat, 11 Oct 2025 21:41:46 +0000

https://mention.com/en/

New comment by prats226 in "Launch HN: Extend (YC W23) – Turn your messiest documents into data"

prats226 — Sat, 11 Oct 2025 19:14:58 +0000

Here is link to open source model: https://huggingface.co/nanonets/Nanonets-OCR-s

And hosted model: https://docstrange.nanonets.com/

New comment by prats226 in "Designing agentic loops"

prats226 — Tue, 30 Sep 2025 21:10:54 +0000

It boils down to information loss in compaction driven by LLM's. Either you could carefully design tools that only give compacted output with high information density so models have to auto-compact or organize information only once in a while which eventually is going to be lossy.

Or you just give loads of information without thinking much about it, assuming models will have to do frequent compaction and memory organization and hope its not super lossy.

New comment by prats226 in "Designing agentic loops"

prats226 — Tue, 30 Sep 2025 21:07:21 +0000

Reason I felt like they are closely connected are because for designing tools for lets say coding agents, you have to be thoughful of context engineering.

Eg linear MCP is notorious for giving large JSONs which quickly fill up context and hard for model to understand. So tools need to be designed slightly differently for agents keeping context engineering in mind compared to how you design them for humans.

Context engineering feels like more central and first-principle approach of designing tools, agent loops.

New comment by prats226 in "Designing agentic loops"

prats226 — Tue, 30 Sep 2025 19:43:35 +0000

Context engineering is another name people have given to same skill?

New comment by prats226 in "Show HN: HumanAlarm – Real people knock on your door to wake you up"

prats226 — Wed, 10 Sep 2025 22:10:05 +0000

You can always put automation for your google home to blast music at full volume at right time. And if you don't wake up from sound of music yourself, your neighbour will knock on your door for sure!

New comment by prats226 in "In a first, Google has released data on how much energy an AI prompt uses"

prats226 — Thu, 21 Aug 2025 21:38:57 +0000

With google serving AI overviews, now an average search query should cost more? Compute is getting cheaper but also algorithms getting more and more complex, increasing compute?

New comment by prats226 in "Training language models to be warm and empathetic makes them less reliable"

prats226 — Tue, 12 Aug 2025 18:51:36 +0000

Read long time ago that even SFT for conversations vs base model for autocomplete reduces intelligence, increases perplexity