Pulse AI

2026 50

1. 28 OctBest PracticesWhy 99 Percent Field Accuracy Still Fails the Financial StatementA 99 percent field accuracy score means a 300-value financial statement comes back fully correct only about 5 percent of the time, which is why Pulse measures extraction by how many documents are entirely right, with ever
2. 27 OctBest PracticesThe Data Room Problem: Classifying Mixed Scans Before ExtractionReal estate data rooms bury leases, amendments, estoppels and rent rolls inside merged, decades-old scans, so Pulse classifies every page and splits each file into its individual documents first, then runs extraction with
3. 23 OctBest PracticesThe Review Queue: Where Extraction Becomes Trusted Financial DataEvery financial document pipeline that makes it to production has a review step in front of the database, and that step, not the model, is usually where projects stall.Sid and RitvikOctober 23, 2026
4. 21 OctBest PracticesChart Reconstruction Recovers Subsurface Data for Pulse’s Energy PartnersSince launch, Agentic Chart Reconstruction has been retrained on degraded multi-curve well logs and now exports depth-indexed LAS files, letting Pulse's energy partners recover decades of legacy subsurface data fr
5. 19 OctBest PracticesReading the Rate Case: Why Utility Filings Break ExtractionUtility rate case exhibits were built for a hearing room rather than a database, and this post walks through why their layout breaks conventional extraction and how Pulse recovers the full table structure with a calibrated conf
6. 7 OctAnnouncementsFrom Documents to Data in SnowflakeMost of what regulated companies know is locked in PDFs, scans, drawings, and maps their warehouse cannot read. Here is how Pulse turns those documents into structured, confidence-scored data that lands in Snowflake as governed tables. Sid and RitvikOc
7. 2 OctAI TrendsPulse + GDP.xlsx: Better Data, Better AgentsHow Pulse extracts spreadsheet evidence for AI agents, and what it showed on Surge's GDP.xlsx benchmark. Sid and RitvikOctober 2, 2026
8. 14 SepBest PracticesWhy Commercial Lending Packages Need Reconciliation, Not Just ExtractionExtraction gets the numbers out of a commercial loan package; reconciliation across the rent roll, operating statement, leases, and appraisal is what tells the underwriter whether they agree.Sid and RitvikSeptember
9. 13 AugAnnouncementsPulse is now on ClaudePulse is now a connector in Claude's directory. Point Claude at any document by URL and get back clean, structured data without leaving the conversation. Sid and RitvikAugust 13, 2026
10. 12 AugBest PracticesThe Screening Matrix: Why Basin Assessment Tables Defeat Ordinary ExtractionBasin screening matrices invert the usual table shape, with rows as evaluation criteria and columns as assets, and most cells carry qualified language rather than clean values. Preserving that structure is what
11. 3 AugBest PracticesWhy Invoice Queues Never Reach Straight-Through ProcessingInvoice capture stalls short of straight-through processing when line items break and routing data sits off the invoice. Calibrated confidence opens the bypass lane.Sid and RitvikAugust 3, 2026
12. 13 JulBest PracticesWhy utility documents break standard extraction pipelinesThe documents utilities generate, collect, and process every day were never built for downstream systems, which is why generic extraction fails on them. Sid and RitvikJuly 13, 2026
13. 8 JunAnnouncementsDocument Metadata Is Now Available in PulsePulse now returns a file's native metadata, including authorship, revision history, timestamps, hidden worksheets, EXIF fields, and hashes, alongside the extracted content in the same Extract call.Sid and RitvikJune 8, 2026
14. 15 MayAnnouncementsPulse Releases Agentic Chart ReconstructionA chart stores most of its meaning in the height of a bar or the slope of a line rather than in any character on the page. Pulse Agentic Chart Reconstruction recovers the values, scales, series, and visual encodings inside it.Sid and RitvikMay
15. 13 MayBest PracticesWhy Financial Spreading Is Still a Manual JobFinancial spreading resists automation because statement meaning lives in visual hierarchy that OCR flattens. Structure-preserving extraction turns transcription into review.Sid and RitvikMay 13, 2026
16. 12 MayAnnouncementsIntroducing ClassifyClassify identifies document types before extraction and routes each file into the right pipeline automatically, cutting cost and removing manual triage from intake. Sid and RitvikMay 12, 2026
17. 11 MayAnnouncementsWelcome to the Agent Economy: Agentic Payments Go Live on PulsePoint an agent at our API with no account, no API key, and no signup, and it can pay for extraction on its own, one call at a time. Sid and RitvikMay 11, 2026
18. 4 MayAnnouncementsIntroducing the Pulse CLIThe Pulse CLI brings Pulse's full document extraction pipeline to your terminal, letting you convert PDFs, folders, and URLs into clean Markdown and structured JSON with a single command. Built for developers, it integrates seamlessly with shell scripts, CI/CD,
19. 1 MayAnnouncementsCapturing what the pen says: Handwritten Mark Detection in PulsePulse is launching Selection Mark, a purpose-built model that reads the checkmarks, circles, underlines, and strikethroughs people make on forms, where frontier VLMs reliably fail. It adds selection state to the standard ex
20. 30 AprAnnouncementsThe Pulse MCP Server Is Now LiveThe Pulse MCP server is live today. Point any MCP client at Pulse, whether that's Claude, Codex, VS Code, or one you've built yourself, and your agent can call extraction, schema generation, splitting, and table parsing directly as tools, deciding for its
21. 29 AprAI TrendsGDP.pdf and the Retrieval Layer: How Pulse Improves Enterprise Document AISurge's new benchmark shows why frontier models need structured, page-grounded evidence to answer questions over professional PDFs. Sid and RitvikApril 29, 2026
22. 28 AprBest PracticesEvaluating Mistral OCR 4 on PulseBench-TabMistral released OCR 4 - the obvious question on our end working in the space was a narrow but vital one: how does it handle tables? Table extraction is one of the toughest parts of document understanding, so we ran OCR 4 through the full Pulse
23. 24 AprAI TrendsThe Difference Between a Digitized Well Log and a Usable OneOperators digitized their well logs years ago, but a scan only preserves the image of a document, not the depth tracks, curve scales, and spatial relationships that make the subsurface record usable as data. This piece shows why th
24. 23 AprBest PracticesWhy Field Tickets Break Cost Reconciliation in Oil and GasField tickets are the financial source records connecting oilfield work to the ledger, yet their endless vendor-by-vendor variation, handwriting, and degraded scans break conventional extraction and quietly corrupt cost reconcil
25. 15 AprBest PracticesHow Subsurface Maps Actually Encode InformationOil and gas is an industry that runs on documents that were never meant for machines, and the capital allocation decisions that matter at an integrated operator are made on the basis of a document stack that spans every era of the business
26. 14 AprBest PracticesThe Documents Upstream Operators Run OnThe documents that drive upstream decisions, from well schematics to stratigraphic charts, hold their meaning in depth columns and spatial layouts that off-the-shelf AI simply can't follow. Pulse reads them the way your engineers do, and delivers
27. 13 AprBest PracticesThe Hardest Problems in Excel ExtractionExcel extraction is widely treated as the easy cousin of PDF extraction. Spreadsheets are structured, the reasoning goes, so a parser should be able to walk the grid, read the values, and emit clean tabular outputSid and RitvikApril 13, 2026
28. 10 AprAnnouncementsIntroducing Pulse Form Fill: Programmatic PDF Editing Pulse Form Fill builds on top of our state-of-the-art extraction models, leveraging the same visual document understanding that powers Pulse's extraction pipeline as the foundation for accurately placing data back into any PDF. Sid a
29. 9 AprBest PracticesDocument Intelligence Built for Regulated WorkflowsPulse turns the highest-value documents in regulated industries, from handwritten geotechnical boring logs to polished institutional research, into structured, audit-grade data with every field grounded back to its location on the sour
30. 8 AprAI TrendsWhere Table Extraction Breaks: Lessons from PulseBench-TabMost table extraction benchmarks test on clean, well-bordered, single-language grids, the kind of tables that look great in a demo but rarely show up in the documents enterprises actually need to process.Sid and RitvikApril 8, 2026
31. 7 AprAI TrendsRD-TableBench Updated Results (2026): Pulse Ultra 2 Leads Across the BoardWe ran Pulse Ultra 2 through RD-TableBench using Reducto's published ground truth and their published Needleman-Wunsch scoring methodology, with no modifications and no corrections, on the same evaluation pipeline Red
32. 6 AprAI TrendsWhy LLMs Cannot Own Enterprise Document ParsingWhy LLMs alone can't handle enterprise document parsing, and what a production-grade ingestion stack actually requires.Sid and RitvikApril 6, 2026
33. 3 AprBest PracticesTable Extraction Benchmarks: What They Measure, What They Miss, and What the Field NeedsA close look at the four most-cited table extraction benchmarks, what they measure, where they fall short, and what enterprise document intelligence actually requires. Sid and RitvikApril 3, 2026
34. 2 AprAnnouncementsIntroducing PulseBench-Tab: A Frontier Benchmark for Table ParsingWe’re announcing PulseBench-Tab, a frontier open-source benchmark focused on table extraction.Sid and RitvikApril 2, 2026
35. 31 MarAnnouncementsIntroducing Pulse Ultra 2 Pulse Ultra 2 is our highest-accuracy document extraction model, built on a rearchitected end-to-end pipeline with native merged cell handling, multilingual OCR across 100+ languages, and improved table boundary detection.Sid and RitvikMarch 31, 2026
36. 27 MarBest PracticesWhy Document Extraction Is a Constraint Satisfaction Problem, Not a Sequence ProblemRegulated document extraction is a constraint satisfaction problem, not a sequence problem, and the schema belongs inside the decoder rather than downstream of it. Sid and RitvikMarch 27, 2026
37. 24 MarBest PracticesThe Document Problem Hiding in Plain Sight Across Regulated IndustriesCritical business data in regulated industries is trapped inside complex documents that defeat conventional extraction tools. Pulse AI's vision-language models process these documents with structural understanding, t
38. 23 MarBest PracticesThe Silent Failures in Financial Document ExtractionFinancial documents encode critical meaning through formatting, inconsistencies, and edge cases that most extraction systems miss. The real challenge isn’t basic OCR or tables, but handling these subtle, real-world complexities withou
39. 20 MarAnnouncementsFootnote Extraction Is Now Generally Available on the Pulse PlatformAutomatically extract and structure footnotes from complex documents with preserved reference links, positional metadata, and full text, making critical fine print from filings, contracts, and reports programmatically a
40. 19 MarAnnouncementsSpanned Table Merging Is Now Generally AvailableSpanned table merging is now generally available through the Cloud API, automatically detecting and combining tables that continue across multiple pages into a single unified table so downstream systems receive complete, structured data wi
41. 17 MarAnnouncementsWord and Cell Level Bounding Boxes Are Now Generally AvailableWord and cell level bounding boxes are now publicly available through the Pulse API, giving every extracted word pixel-precise coordinates and every table cell its exact geometric boundaries, including merged cells, multi-lin
42. 13 MarBest PracticesWhy Rent Rolls Are One of the Hardest Documents in Commercial Real EstateRent rolls are among the hardest documents in commercial real estate because their dense, multi-page tables, reliance on blank cells for meaning, and highly variable formats break traditional OCR, requiring layout
43. 12 MarBest PracticesHow Banks Are Rethinking Document Infrastructure as LLMs Enter the WorkflowBanks are realizing that LLMs do not unlock value on their own, and that the real bottleneck is transforming complex financial documents into accurate, structured, and usable data that can power high stakes work
44. 11 MarAnnouncementsSplit Endpoint Is Now Generally AvailableDefine and isolate the exact pages that matter before extraction, reducing noise and improving accuracy across your document pipelines.Sid and RitvikMarch 11, 2026
45. 10 MarBest PracticesFrom 10-K to Dataset: Automating Financial Research PipelinesTurn complex 10-K filings into auditable, structured datasets with layout-aware extraction, schema mapping, and full source traceability.Sid and RitvikMarch 10, 2026
46. 6 MarBest PracticesWhy Word-Level Bounding Boxes Are Non-Negotiable for Enterprise Data PipelinesPulse goes beyond text accuracy by preserving word-level spatial context for every extracted value. This enables precise source linking, reliable audits, and production-ready document AI you can trust.Sid and
47. 5 MarBest PracticesConfidence Scoring for Downstream Automation: When to Trust, When to ReviewWhy most AI confidence scores are misleading and how Pulse calibrates them to real-world accuracy, enabling reliable automation, smarter review workflows, and predictable error rates in production.Sid and Ritvik
48. 4 MarBest PracticesHow Fountain Cut Document Processing Time by 90% with PulseFountain partnered with Pulse to power its Compliance engine, cutting document processing time by 90 percent while improving extraction accuracy and scalability across industries.Sid and RitvikMarch 4, 2026
49. 2 MarBest PracticesProcessing at Scale: What Changes When You Go from 1,000 to 1 Billion PagesA behind-the-scenes look at how Pulse engineers its document pipeline to handle bursty, multi-billion-page workloads with high reliability, low latency, and efficient use of compute.Sid and RitvikMarch 2, 2026
50. 24 FebAI TrendsMerged Cells, Broken Data: How Span Tables Defeat Document ExtractionSpan tables with merged cells are one of the biggest hidden failure points in document extraction. They encode hierarchy visually rather than explicitly, which causes traditional tools to misassign values to the wrong head