After processing 400M+ pages for the world's largest investment firms, AI startups, and Fortune 500s, Pulse is launching Ultra: their new hybrid reasoning model. It's the most accurate document extraction model in the industry. Live for all customers today.

After processing nearly 500M pages, we discovered the biggest challenge in document AI isn't OCR accuracy - it's semantic understanding across page breaks and column boundaries. 🧵 (1/8)

Pulse (Pulse) has just launched Meridian, an AI-powered financial document processor that can automatically convert any PDF, Word doc, PowerPoint presentation, or image into a structured Excel export with charts and graphs. runpulse.com/blog/introduci… Congrats on the launch,

we're super excited to be launching Meridian publicly! no more analysts having to manually copy numbers from pdfs into spreadsheets at 2 am before a board deadline if you're interested in trying it out give me a DM! 🫡

the team at Pulse put bytedance's dolphin OCR to the test against complex documents that matter for real business use cases. while it shows improvements in reading order detection, we found critical limitations across key areas: - 7.7% structured data extraction from

@pulse__ai team just dropped why "98% accurate" document extraction still breaks in production with 4000 errors per 1000 pages. single accuracy scores miss broken reading order, shifted table columns, and lost cross page context that silently corrupt entire datasets. we’ve

Pulse (Pulse) just launched their state-of-the-art document extraction platform. It turns complex PDFs, scans, decks, and images into LLM-ready data. No training required. runpulse.com/blog/pulse-ope… Congrats on the launch, sid and Ritvik Pandey!

customer feedback last week hours of work → minutes with @pulse__ai

Join us!

Almost a year since this

@pulse__ai just launched formula recognition. trained on 10m+ formula/latex pairs from papers + handwritten notes. traditional ocr breaks on math (α, β, fractions, matrices). our model treats formulas as structured objects → clean latex. built on pulse’s production-grade

Culture building is everything when you're asking engineers to solve the hardest problems for enterprises. The entire Pulse team is usually in the office 12 hours a day - everyone needs to be in one place, building together. Having an immediate feedback loop is incredibly

The Pulse team just published "The Precision Tax" - why "99% accuracy" fails in finance. One percent error in financial document processing means broken valuations, failed covenant tests, and regulatory exposure. The real benchmark isn't accuracy, it's determinism. Same

Pulse is now officially part of Cloudera's Enterprise AI ecosystem. Excited to partner with Cloudera and continue delivering the most accurate document extraction models at enterprise scale.

🧢

.Pulse just launched Ultra Nano, their new enterprise-focused document extraction model with complete self-hosting, already running across Fortune 50s, insurers, investment firms, banks, and foundational model labs. runpulse.com/blog/self-host… Congrats on the launch, sid

DeepSeek AI dropped a new open-source OCR model today 👀 At @pulse__ai, we tested it on financial docs, handwritten forms, and complex tables. The results showed the same issues plaguing LLM-driven OCR: - Unstable outputs - Hallucinated text - Broken table structures Reality

threw a screenshot of this post into Pulse ~99% accurate try it here: platform.runpulse.com

Exciting research preview to share on XLSX parsing at Pulse . Spreadsheets are deceptively hard - merged cells, multi-tab workbooks, and cross-sheet references break when you flatten them. Our team has developed and implemented a token-efficient encoder resulting in

spotted