> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dataframer.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# What is DataFramer?

> DataFramer unifies AI traces, user behavior, and expert judgment into one loop to measure and improve the accuracy and value of your AI workflows.

DataFramer is an AI Workflow Intelligence Platform for accurate, high-value AI workflows. It helps answer the following questions:

1. Is the AI behaving accurately within the context of your business?
2. Is the complete AI-powered workflow delivering value to your users and the organization?

It does that by bringing together AI traces, user behavior, worflow events, and expert judgment into one repeatable operating layer. It connects individual model and agent interactions to the user journeys and workflows in which they occur.

From there, DataFramer turns this data into a continuous cycle of analysis, human review, evaluation, and measurable workflow improvement.

## How it works

DataFramer follows one loop, end to end:

**Capture & Correlate**

Capture user interactions, feedback, and workflow events, from your applications with the [Signals SDK](/user-signals/overview), [connect your AI observability tool](/user-signals/connect-langfuse-langsmith) (ex. Langfuse, LangSmith) to pull existing traces with metadata, and auto-tag traces with [`dataframer-journey`](/user-signals/server-instrumentation). DataFramer stitches signals, events, and traces together with a shared journey id, so a user's actions and the AI traces around them can be unified into User Journeys.

**Measure User Journeys**

See the complete timeline of each [user journey](/journeys/overview) or their workflow, including user actions, AI traces, agents, and workflow events, in one place. Then [zoom out to analyze](/journeys/dashboards) all journeys and understand completion, adoption, drop-offs, throughput, and cycle time to understand where AI is helping or hurting the workflow and prioritize improvements that increase accuracy, efficiency, and business value.

**Discover & diagnose patterns**

[Findings](/findings/overview) lets you search your traces for any AI behavior you care about, not just failures or accuracy problems. It surfaces the known patterns you already track and helps you find new ones. From a pattern you can look into likely causes, focus on the deviations that matter most, watch how often each one comes back, and check that a fix really worked without breaking something else.

**Human review operations**

Send selected traces to human experts in [Human Reviews](/reviews/overview), the structured layer that captures expert judgment as reusable data, not scattered feedback. Grading standards are set in [Rubric Studio](/reviews/rubric-studio) and humans grade traces through [the operational loop](/reviews/operational-loop), with the [Review Copilot](/reviews/copilot) speeding them up. Every correction becomes ground truth for your judges and datasets, reused on the next similar case instead of redone from scratch.

**Evaluate & generate**

Build a [Judge](/judges/overview) from the same grading standards and calibrate it against your reviewers' verdicts until its scores align with theirs, then run it across [datasets of real reviewed traces](/judges/datasets-and-evaluations) to score quality at scale and catch regressions before they reach users. For gaps your traces don't cover, [generate synthetic data](/concepts) grounded in your real data.

# Everything compounds

Each step feeds the next, and the loop closes. [Journeys](/journeys/overview) show where AI helps or hurts. [Findings](/findings/overview) turns that into patterns, [Reviews](/reviews/overview) turns patterns into expert judgment, and [Judges](/judges/overview) and [datasets](/judges/datasets-and-evaluations) turn that judgment into quality you can trust at scale, which sharpens the next round.

Along the way, context builds up. Your corrections, the rubrics and examples that define what good looks like, and the failure patterns you name all add up into a shared memory of how your business judges quality. New reviews start from judgments already made, judges calibrate against more and more ground truth, and a new workflow starts from the standards you've already set instead of from zero.

**Measured in business terms**

All of it rolls up across your [journeys dashboards](/journeys/dashboards). You see the outcomes that matter, like completion, drop-off, throughput, cost, and cycle time, so you can tell whether AI is delivering value, not just running.

<CardGroup cols={2}>
  <Card title="Signals & Journeys" icon="route" href="/user-signals/overview">
    Capture user events and correlate them with AI traces
  </Card>

  <Card title="Findings" icon="magnifying-glass" href="/findings/overview">
    A multi-stage system that surfaces the real issues in your traces, over 82% accurate
  </Card>

  <Card title="Human Reviews" icon="user-check" href="/reviews/overview">
    Structure expert review and reuse feedback
  </Card>

  <Card title="Judges & Datasets" icon="scale-balanced" href="/judges/overview">
    Calibrate judges and generate synthetic data
  </Card>
</CardGroup>

## Also: synthetic data generation

Separately from the accuracy loop above, DataFramer can generate realistic, diverse synthetic datasets at scale, from example data or a text description alone. See [Core Concepts](/concepts) and the [Quickstart](/quickstart).

## Programmatic access

<Card title="API & MCP" icon="code" href="/api-and-mcp">
  Python SDK and MCP server for datasets, specs, generation runs, and evaluations. Findings, Reviews, and Judges are UI-only today: no public API yet.
</Card>