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

---
title: Worked examples by outcome
description: >-
  One worked example per customer outcome, each with a tiny fixture, a quote, a
  separate approval and a real result bundle. Filter by your role and industry
  instead of hunting for a sector copy.
route: /docs/compute/examples
page_id: docs-compute-examples
page_type: hubIndex
content_layer: guided documentation
surface_profile: guided_docs
audience: People choosing which Compute walkthrough to run first
voice: D
reader_question: Which worked example matches the outcome I want?
primary_action: Open the example closest to your outcome
source_locale: en-US
source_status: APPROVED
source_version: docs-compute-examples-en-us-2026-09-05-v1
last_updated: '2026-09-05'
robots: index
claim_registry_pinned: true
claim_registry_resync: "npm run governed:tsx -- scripts/build-compute-public-claim-registry.ts --write"
---

## Pick by outcome, not by industry

Each example below is one outcome, run end to end on a fixture small enough to read by hand. They all follow the same five steps: inspect the input locally, read a quote, approve it separately, submit, then read the bundle.

There is one example per outcome, not one per sector. A legal team, a hospital group and a media publisher preparing the same kind of input run the same walkthrough. What changes is your data profile, your workflow parameters and who approves the spend. So treat role and industry as filters over one canonical example, and skip the ones that do not describe your outcome.

Every example stops before an outbound action. Compute prepares work and returns evidence with it. A person reads that evidence and decides what to send, publish or commit.

## The examples

- [Build a searchable media archive](/docs/compute/examples/media-archive) — timestamped transcripts, scenes and a segment index from a folder of recordings.
- [Turn meeting recordings into cited actions](/docs/compute/examples/meeting-intelligence) — a verbatim transcript plus action drafts, each carrying the line it came from.
- [Measure call quality without one blunt score](/docs/compute/examples/call-quality) — talk time, interruptions and issue flags reported separately, with evidence.
- [Prepare podcast and video episodes](/docs/compute/examples/podcast-video-factory) — transcript, chapters, clip candidates and draft copy, each with its own readiness.
- [Translate a corpus with a pinned glossary](/docs/compute/examples/translation) — one language pair, aligned evidence and an explicit review path for ambiguity.
- [Extract text, tables and page evidence](/docs/compute/examples/document-intelligence) — a narrowing schema tested against scanned and rejected pages.
- [Normalize and deduplicate a dataset](/docs/compute/examples/data-preparation) — stable item identity through every step, with accepted and rejected rows exported apart.
- [Compare candidates against a held-out set](/docs/compute/examples/evaluation-lab) — dataset, candidate and rubric kept apart, with slice metrics and the disagreements.
- [Let your AI agent run a batch and read typed results](/docs/compute/examples/agent-workbench) — typed rows an agent can act on, with the quote and approval gates in the loop.
- [Review code with bounded evidence and real test output](/docs/compute/examples/code-quality) — suggested findings kept separate from tests that actually ran.
- [Run a test suite and collect the evidence](/docs/compute/examples/apple-test-lab) — per-case outcomes, logs and artifacts, with a red suite distinguished from one that never compiled.
- [Run an approved Core ML model over a batch](/docs/compute/examples/coreml-lab) — real predictions and measured performance, never a training claim.
- [Run vision observations across an image set](/docs/compute/examples/vision-batch) — declared observations with image evidence, uncertainty and privacy limits stated.
- [Turn a document corpus into cited chunks](/docs/compute/examples/corpus-to-rag) — headings, pages and citations preserved, with no promise about retrieval quality.
- [Check claims against a bounded source set](/docs/compute/examples/evidence-review) — unresolved and conflicting items preserved rather than silently settled.