pierre

Liteparse now support markdown, beat markitdown in speed and accuracy, fully open source https://github.com/run-llama/liteparse

Yes, it evaluate using frontier model for parsing from all 3 major provider (google, anthropic and openai). It is also easy to extend to evaluaye new model (code/dataset is available)

Build a benchmark to evaluate how good document parser work on a dataset of 2000 PDFs manually annotated, trying to evaluate accross multiple dimensions: charts, tables, text styling, text correctness, and attribution.

The benchmark evaluate performance on full page (not selected part of the pages), and evaluaye different OSS / crobtier model / commercial approach.

For transparency it is available as a HF leaderbaord.

Paper: https://arxiv.org/abs/2604.08538

A fast and open source spatial text parser

A new document (including PDF) parser that outperform traditional tool such as PyPDF or MuTools.

Link to open source repo: https://github.com/run-llama/liteparse

Contributor here, happy to answer any questions!

Demo: https://olmocr.allenai.org/ Paper: https://arxiv.org/abs/2510.19817

Main issue is that token are not equivalent across provider / models. With huge disparity inside provider beyond the tokenizer model:

- An image will take 10x token on gpt-4o-mini vs gpt-4.

- On gemini 2.5 pro output token are token except if you are using structure output, then all character are count as a token each for billing.

- ...

Having the price per token is nice, but what is really needed is to know how much a given query / answer will cost you, as not all token are equals.

LlamaIndex | Senior/Staff Software Engineer (LlamaParse) | San-Francisco, CA | Remote | Full-time | $100K – $300K + Equity | https://www.llamaindex.ai/careers

LlamaIndex is building a platform for AI agents that can find information, synthesize insights, generate reports, and take actions over the most complex enterprise data.

We are seeking an exceptional engineer to join our growing LlamaParse team. Will work at the intersection of document processing, machine learning, and software engineering to push the boundaries of what's possible in document understanding. As a key member of a focused team, will have significant impact on our product's direction and technical architecture.

We are also hiring for a range of other roles, see our career page:

- Backend Software Engineer

- Forward Deploy Engineer

- Founding AI Engineer

- Open Source Engineer Python

- Founding Lead Product Manager

- Platform Engineer

- Senior Developer Relation Engineer

- Senior / Staff Backend Engineer

- Product Marketing Manager

If you want to try agentic parsing we added support for sonnet-3.7 agentic parse and gemini 2.0 in llamaParse. cloud.llamaindex.ai/parse (select advanced options / parse with agent then a model)

However this come at a high cost in token and latency, but result in way better parse quality. Hopefully with new model this can be improved.

I read the output header, and see the Artifex (mutools / gs team) headers

This is a nice UI for end users, however it seems to be a seems wrapper on top of mutool, which is distributed as AGPL. If you want to process PDF locally, legally and safely you should use their CLI instead.

Parsing docs using LVM is the way forward (also see OCR2 paper released last week, people are having ablot of success parsing with fine tunned Qwen2).

The hard part is to prevent the model ignoring some part of the page and halucinations (see some of the gpt4o sample here like the xanax notice:https://www.llamaindex.ai/blog/introducing-llamaparse-premiu...)

However this model will get better and we may soon have a good pdf to md model.

yes, you can pass an array of path to the extract function.

Good article, but what is the alternative? What can you build today as a software engineer that can have impact? Nothing seems to come close to AI / AI infra, even of its hard / risky / a moving landscape.

Locally you can choose pypdf or mupdf wich are good but not perfect. If you can send your data online llamaparse is quite good.

RAG cli from llamaindex, allow you to do it 100% locally when used with ollama or llamacpp instead of OpenAI.

https://docs.llamaindex.ai/en/stable/getting_started/starter...

For PPT, chuncking 'per page' work often quite well. With LlamaParse this will mean splitting on the "\n---\n" page separator token.

Performance depend on the language / type of docs. Main reason for contemplating switching is that easyOCR seems to not be maintained anymore (no commit in the repo in last 5 months)

Yes, however we will soon support other filetypes natively, and this will lead to better results (when converting from one format to another, there is often some information loss)

EasyOCR, may switch to paddleOCR in the future.

I'm part of the team that build LlamaParse. It's net improvement compare to other PDF->Structured Text extractors (I build several in the past, includig https://github.com/axa-group/Parsr).

For character extraction, LlamaParse use a mixture of OCR / character extraction from the PDF (it's the only parser I'm aware of that address some of the buggy PDF font issues, check the 'text' mode to see raw document before reconstruction), use a mixture of heuristic and Machine learning models to reconstruct the document.

Once plug with a Recursive retrieval strategy, allow you to get Sota result on question answering over complexe text (see notebook: https://github.com/run-llama/llama_parse/blob/main/examples/...).

AMA

In the meta store :)

Went there, was the only customer for the hour of demo I had.

It's sort of happening since 2 year already in china internet courts (for small internet vendor dispute) in Beijing and Hangzhou.

In english: https://iacajournal.org/articles/10.36745/ijca.367

Most of the hole-in-one cover payout do not happen for contest, but for individual that are insured against a hole in one. (source: I work for an insurer)

It is a common provision on the kind of insurance that come attached to a credit card, and you probably are cover for it if you own a premium card in the US or Europe. It generally come with different variation: only on registered games / all games, fix cash amount or expense of the drinks at the club, ...

A common exclusion is that you can not own a professional golf licence.

A few examples: A British Bank Travel insurance (page 33) https://assets.ctfassets.net/s0jgb0x75qln/413D8sFB4gVNoNWZXn...

Amex (page 85) https://americanexpress.com/content/dam/amex/za/network/docu...

If your are interested, the Swiss federation just launch a call for startup / company to express interest for use of U-Space. As I understand it, an authorization to use Swiss U-Space will allow a company to automatically be eligible for whole UE airspace in the future.

Regarding product comparaison it is possible to build a computable model of your policy and the major competitor, and from there automatically benchmark them to find key difference in coverage (what is the maximum delta) or running them through a set predefined claim scenario. It will however require ~3-5 day of work per policy to build the models. Beyond the marketing effort this models can also be reused for risk management / claim management / leakage prevention if implemented right.

Thanks for your answer. I'm not American so not really familiar with your health system. Agreed that B2B will be easier to scale given acquisition cost. However scaling with corporate client limit you in your ability to do good risk selection as I believe you will have to accept all employees.

First of all congrats on the launch, we need more innovation in the space and I applaud your efforts. Here are some thoughts / questions if you want to answer them

We have [...] benefits that exceed the best plans from legacy insurers.

You claim that your product is better than some of the competition, but do not demonstrate it (here or on your website). At this point it seems to me that every insurer claim that they have the best plan with no way for me to know easily.

I think the key issue of the industry as a whole is that there is no way for client to compare insurance product beyond pricing. A product that could look better at covering lenses for example because it cover up to $2000/year vs another that cover up to $350 may actually be worse because of some widely applicable exclusion written into the contract.

What are your thoughts on this point?

On top of that, we reduce premiums by up to 20% for employers. We can do this because we are a software company that owns an insurance carrier. We automate roughly half the tasks involved with claims, care coordination, underwriting and back-office operations. We aggregate data from disparate sources (claims, clinical, pharma, lab, and wellness data) to make superior decisions and aid patients. Our technology helps members identify and treat conditions earlier and more effectively. We also have a much better user experience—a single portal to access telehealth, care concierge, claims data, wellness plan, doctor lookup, rewards card, etc.

You claim to be able to reduce cost because of your tech, and I believe compare to legacy carrier, your IT / Process are cheaper to run today (I believe that legacy carrier spend ~5-10% of their revenue on IT). From my observation the insurance industry is quite bad at getting ride of legacy systems (for compliance, once you decommission a system you sometime need to prove that the new system run the old policy the same way, or just because to many process optimisation software has been build on top of the legacy system making it extremely costly to sunset). How do you plan to maintain this cost down once you extend to new states / product /over time, to keep this cost advantage?

Since half of Americans get health coverage through their employer, we’re focused on companies to maximize impact.

I understand that B2B distribution is easier than B2C, but this can go against your mission of changing healthcare incentives for mutual benefits. You customer are the Employers, and their incentives are to reduce cost and to maintain their employee healthy short term, whereas employee would like to have better access to healthcare (higher cost) and to stay healthy Long term. How will you find balance here? What happen when a a major client as you to cut cost for their plan to the expense of the employee coverage and you need to keep them as a client to keep the company afloat.

Minister of the future intro about a heat bubble gave me nightmares.

It make me think about how we could prevent death in case of a massive humid heat wave, and how we could do it at a low enough cost, especially in a situation where the grid fail. My best guess will be some tent with high isolation fabric (think emergency blanket) cooled with a tank of liquid nitrogen.