Maia

Data labeling and quality platform for training and evaluating AI models.

Kili Technology is a data-centric platform for turning raw data into high-quality training and evaluation datasets. It supports annotation across image, video, text, OCR, and geospatial data, with review and quality workflows, plus LLM evaluation and RLHF using human-in-the-loop and LLM-as-a-judge. It is used by enterprises including Airbus and SAP, and offers cloud, private-cloud, and on-premise deployment.

Production-grade extraction for complex documents.

A document-extraction platform that converts messy, real-world documents — financial statements, medical records, contracts, spreadsheets — into clean, LLM-ready structured data. Pulse runs its own OCR, layout, and vision models (including its Ultra extraction model) rather than wrapping a general-purpose LLM, and exposes the pipeline through an API that drops into existing data workflows. It offers a free sandbox to try, with enterprise tiers for scale; the company says it has processed over a billion document pages.

Turn unstructured documents into structured data.

An agentic document-processing platform that extracts clean, structured JSON from PDFs, scans, and other complex documents using LLMs. Its Prompt Studio gives a no-code IDE to author and test extraction prompts per field, which you then deploy as APIs or ETL pipelines into your warehouse. Built by Zipstack, Unstract is open source under AGPL-3.0 and self-hostable via Docker Compose, with a managed cloud that adds SSO, human-in-the-loop review, and compliance certifications (SOC 2, HIPAA, ISO 27001, GDPR).

Git extension for versioning data, models, and ML experiments.

DVC (Data Version Control) brings software-engineering practices to machine learning: it versions datasets, models, and pipelines alongside code in any Git repository, storing large files in your own remote storage while keeping lightweight pointers in Git. Enables reproducible experiments, data/model lineage, and pipeline orchestration from the command line.

Git-like version control for data lakes over your existing object storage.

Open-source data version control that turns object storage (S3, GCS, Azure Blob, MinIO) into Git-like repositories. Teams branch, commit, merge, and roll back petabyte-scale data lakes for isolated experimentation, reproducible ML pipelines, data-quality gates, and compliance lineage — without copying data. Integrates with Spark, Trino, Databricks, Delta Lake, and Iceberg.

High-accuracy document parsing — PDFs and images to markdown, JSON, and HTML.

Datalab turns PDFs, images, and office documents into clean markdown, JSON, and HTML with layout, table, math, and code preservation. It is the commercial, hosted layer over the open-source Marker converter and Surya OCR toolkit, offered as a pay-as-you-go API with a free monthly allowance, while the underlying models stay free to self-host for research and small startups.

Enterprise AI platform for building, deploying, and governing ML and agentic AI.

DataRobot is an enterprise AI platform spanning the full lifecycle: predictive AI with automated machine learning (AutoML), generative and agentic AI, plus observability and governance. Its AutoML builds and compares many models at once so teams ship production-ready AI faster, and newer agent kits make enterprise AI agents practical to deploy. It runs in the cloud or on-premises for large organizations.

Full-stack document processing platform for AI agents and pipelines.

Extend is an LLM-powered document processing platform that parses, extracts, classifies, splits, and edits complex documents — handwriting, tables, and mixed formats — into reliable structured data via API or its web Studio. It combines multiple frontier models with proprietary context engineering for messy real-world files. Used by teams at Brex, Square, Checkr, and Flatiron Health.

Incremental data framework for fresh AI context.

CocoIndex is an open-source data transformation framework that keeps AI agents and LLM apps supplied with continuously fresh, structured context. It turns sources like codebases, PDFs, databases, and Slack into vector or graph stores, and reprocesses only what changed (delta-only) with parallel execution by default. A Rust core drives reliability while pipelines are defined declaratively in Python, with end-to-end lineage and an observability UI called CocoInsight.

Web-scale data extraction and a knowledge graph that grounds AI in facts.

Diffbot reads the public web like a person and turns it into structured data: an Extract API for articles and products, a Crawl service, a Natural Language API, and a Knowledge Graph of billions of entities and over a trillion facts. The graph is refreshed continuously, so AI systems can ground answers in current, verifiable data rather than model memory. Diffbot also ships its own factually grounded GraphRAG language model.