OuiDire.app • by Studiorium.ai
How OuiDire Works

The Quick Guide shows what to click. This guide explains what the app is doing behind those clicks: workflow, analysis layers, exports, safeguards, costs, and system design.

Quick Guide Version française

How OuiDire Works

The invisible work behind the audit workspace

The Quick Guide shows what to click. This guide explains what the app is doing behind those clicks.

OuiDire is not a record summarizer.

It is an audit workspace for psychiatric and legal records: a way to break dense documents into reviewable units, identify weak claims, separate sources from interpretations, compare human review with machine suggestions, and produce exports that make the record easier to challenge, explain, and discuss.

The visible workspace is only the surface. OuiDire may look dense at first because the problem itself is dense. The work behind the app is to make that density reviewable, traceable, and answerable.

OuiDire's work happens across several layers: OCR, cards, Strategic Overview, inter-document analysis, source review, macro annotation, Gemini suggestions, human validation, Deep Critical Read, Gemini Digest, Master Brief, cost tracking, save/restore, and export architecture. Each layer has a specific role. Together, they turn a dense file into structured material that can be reviewed by a patient, lawyer, researcher, advocate, or careful reader.

OuiDire does not replace lawyers, clinicians, experts, or human judgment. It helps make the record answerable.

1. What OuiDire does

Psychiatric records often appear final because they are long, formal, and written in institutional language. Claims can be repeated across documents, reframed over time, detached from their original source, or treated as settled facts without careful revalidation.

OuiDire helps users slow that process down.

It turns records into cards. Each card becomes a small unit of review: a claim, passage, observation, or excerpt that can be read, tagged, challenged, compared, and later exported.

The goal is not to produce a single AI answer. The goal is to create a structured audit trail.

OuiDire helps users ask:

This is why OuiDire is both an annotation tool and an audit tool. It does not simply annotate records. It helps annotate flaws in psychiatric records.

2. The basic workflow

The basic workflow is:

The important rule is: OCR first. Cards second. AI third.

Until cards exist, there is nothing stable for the app or the user to review. The starting material can be a PDF or usable photos of paper pages; the goal is the same: create readable cards.

Once cards exist, the user can generate a Strategic Overview, work manually, use Gemini suggestions, run Full Analysis: Both Macro Sets, launch a Deep Critical Read, or generate exports.

OuiDire is designed so users can start small. A first session does not need to process an entire file. A useful first session might involve uploading a short PDF or a few photographed pages, creating cards, reviewing 5 to 10 cards, testing Gemini on a small batch, and saving the work.

3. The workspace

OuiDire's workspace is organized around a practical reading cycle.

The left side is intake: documents, cards, OCR, and navigation.

The center is inspection: reading one card carefully, checking source/hearsay, reviewing macros, and deciding what has actually been confirmed.

The right side is audit: notes, findings, machine suggestions, exports, and deeper analytical tools.

The maps and visual layers help users zoom out. They show patterns that are hard to see when reading one card at a time.

The app is not just a form. It is a cockpit for file review.

OuiDire also separates two workspace modes.

Focus Mode helps the user concentrate on the current card and annotation controls. Big View brings back the full workspace: intake, audit, maps, deliverables, and navigation together. The mode is not a different analysis. It is a way to control how much of the workspace is visible while reviewing.

4. OCR and card creation

OCR is the first invisible layer.

Many psychiatric and legal records arrive as PDFs that are not directly readable as structured text. Others arrive first as paper pages, which the user may photograph before importing. OCR turns the visual document, whether PDF scan or usable page photo, into text that can be segmented, reviewed, and cited.

OuiDire's OCR layer matters because the rest of the system depends on it. Bad OCR creates weak cards. Weak cards produce weaker annotations, weaker AI suggestions, and weaker exports.

The purpose of OCR is not only to extract text. It is to create reviewable material from whatever the user actually has: a scanned PDF, an image-based document, or photographed paper pages.

Once OCR has run, OuiDire transforms the document into cards. These cards become the core unit of the audit.

A card is small enough to review carefully, but structured enough to support larger patterns later.

Technical note - OCR fallback

OuiDire's OCR layer is not only a read-the-PDF step. It is a resilience layer. The app can use Google Cloud Document AI for document extraction, and it is designed to support fallback OCR routes when available so difficult PDFs or usable page photos are not automatically lost.

This matters because the entire audit workflow depends on the quality of the cards created from OCR.

5. Strategic Overview

Strategic Overview is the Gemini-powered orientation layer and the first analytical payoff.

It is designed to help the user understand the terrain before reviewing the record card by card. Powered by Gemini, it can generate a thesis, narrative clusters, clinical clusters, key tensions, critical omissions, and inter-document patterns from a capped sample of the record.

This makes it useful at the beginning of a review, especially when a file is long, repetitive, split across several documents, or hard to enter. It helps the user see what may deserve attention first.

The inter-document layer is one of the major gains. Psychiatric files often build force across documents: a claim appears once, is repeated later, changes tone, loses its source, or becomes treated as settled. Overview can help surface those cross-document movements so the user is not trapped in isolated page-by-page reading.

Inter-document analysis can help identify:

But Overview is not the same as a complete analysis.

Its sample is intentionally capped. The right use is: get thesis material and inter-document leads quickly, then verify methodically. The Overview can point toward patterns, tensions, omissions, and cross-document movements; card-by-card review, human annotation, later Gemini suggestions, and Deep Critical Read are the layers that test and refine those signals.

OuiDire currently uses two Overview depths:

Neither should be treated as exhaustive. The Overview is a map of what to inspect next, including what to compare between documents, not final proof.

6. Manual review

Manual review is the foundation of OuiDire.

The human reviewer is the final oracle.

Gemini can suggest. DCR can surface patterns. Master Brief can synthesize. But the human reviewer decides what is actually useful, accurate, misleading, confirmed, rejected, or worth exporting.

Manual review includes:

A card should be marked done only when it has actually been reviewed, not merely seen.

This distinction matters because later exports and metrics become more meaningful when they are based on reviewed material.

Progress should also respect the active macro set. A card reviewed under one macro set should not automatically appear complete under another macro set, because the second reading may ask different questions. This is an important product rule for keeping review state honest.

7. Source and hearsay layer

Source review is separate from macro annotation.

A passage can raise source/hearsay concerns even if it does not fit a macro. A passage can also fit a macro without being hearsay.

This separation is central to OuiDire.

The source layer asks:

The macro layer asks a different question: what kind of flaw or mechanism is visible in the record?

Keeping these layers separate prevents confusion. Hearsay is not just another macro. It is a source/provenance issue.

8. Macro annotation

Macros are OuiDire's mechanism layer.

They help identify recurring patterns in psychiatric records: narrative distortion, recycled history, unsupported extrapolation, diagnostic closure, risk inflation, and other forms of record failure.

A macro is not a final legal conclusion. It is a structured annotation that says: this passage may show a recognizable pattern.

OuiDire currently works with different macro sets because not all files fail in the same way.

One file may be dominated by narrative distortion. Another may be dominated by clinical or diagnostic failure. Another may show both.

Macro sets allow the app to keep the review focused while still supporting broader analysis.

9. Machine suggestions

Gemini can be used to suggest tags, patterns, and findings.

This is not the same as letting AI decide the case.

Gemini's role is to help the reviewer see possible issues faster, especially across long, repetitive, or dense records.

The user can accept, reject, or ignore suggestions.

That distinction is essential.

A machine suggestion is not the same as a human-confirmed finding. OuiDire preserves that difference so exports and metrics can remain honest.

The app's value increases when it can show:

This human/machine distinction is part of the audit trail.

10. Batch vs Document

OuiDire supports different scopes of analysis.

Batch mode is for smaller work: a page range, a small group of cards, or a limited test. It is useful when the user is learning the app, checking OCR quality, or reviewing the file progressively.

Document mode is broader. It is for document-level analysis when the user wants patterns across the whole file.

Inter-document analysis is broader again. It asks what happens across several documents: what repeats, what changes, what contradicts, and what becomes treated as fact over time.

A simple rule:

This matters because larger scopes can cost more and produce more output. Starting small protects both the user and the quality of the review.

11. Full Analysis: Both Macro Sets

Full Analysis: Both Macro Sets lets Gemini read through more than one analytical lens.

Instead of asking the model to look only through the active macro set, Full Analysis: Both Macro Sets can compare multiple macro sets and surface different kinds of findings.

This is useful because a psychiatric record may fail in more than one way.

One passage may show narrative distortion. Another may show diagnostic closure. Another may show risk inflation or recycled prior history.

Full Analysis: Both Macro Sets is designed for richer analysis, not for replacing human review.

The user still needs to evaluate what the machine produced.

12. Deep Critical Read

Deep Critical Read is a deeper analytical layer.

It is designed for more detailed provider-based reading using a richer set of micro-tags and analytical signals.

DCR can surface more precise findings than a basic Gemini call. It can help identify subtle problems, organize observations, and provide stronger material for later synthesis.

DCR is especially useful when the user wants to move beyond simple card tagging and into deeper file analysis.

In OuiDire's current architecture, Deep Critical Read uses OpenAI GPT as a distinct analytical layer. Gemini supports broad document reading and suggestions; GPT supports deeper critical analysis; the human reviewer remains the final layer of validation.

But DCR is still not magic. Its quality depends on the quality of the OCR, the cards, the document scope, and the material available.

DCR findings can strengthen later exports, especially Master Brief, because they give the system more detailed analytical material to synthesize.

13. Gemini Digest

Gemini Digest is granular.

It preserves machine findings card by card.

This is useful when the user wants traceability: not just what did the app conclude, but which card produced which suggestion.

Gemini Digest can help users revisit findings, compare them against the original cards, and decide what deserves human confirmation.

It is not the same as Master Brief.

Gemini Digest is closer to a structured machine-reading log.

Master Brief is a reader-facing synthesis.

14. Master Brief

Master Brief is the synthesis layer.

It takes reviewed material, annotations, metrics, selected findings, and machine outputs, then organizes them into a structured brief.

Its purpose is to help explain the state of the file.

A good Master Brief should not merely list tags. It should help a careful reader understand what is happening in the record.

It may identify dominant patterns, recurring flaws, source problems, repeated narratives, contradictions, or areas where the file appears to rely on weak or recycled material.

Master Brief is strongest when the inputs are strong:

Master Brief should be understood as a synthesis of worked material, not a magical one-click truth machine.

It is useful for:

It should distinguish carefully between human-confirmed findings and machine-generated suggestions.

That distinction protects both the user and the credibility of the export.

15. Save, restore, and export

OuiDire is designed around user control.

Save/restore allows users to keep their work and return later. This matters because file review is slow, cumulative work.

Not every output has the same purpose.

This hierarchy should be clear because different users need different outputs.

A patient may need to understand their file. A lawyer may need a concise state-of-the-file synthesis. A researcher may want patterns. An advocate may need a shareable explanation.

OuiDire should support all of these without pretending they are the same task.

16. Cost Truth

Some actions in OuiDire are simple interface actions. Others require provider compute.

Reading cards, filtering, reviewing, annotating, saving, and restoring are not the same as launching OCR, Gemini, Full Analysis: Both Macro Sets, DCR, or Master Brief.

The Credits section is the user-facing place where credits are added and where the balance connects to compute-heavy actions.

Compute-heavy actions cost money because they call external providers.

OuiDire's cost doctrine is based on visibility rather than mystery.

The user should understand that:

The goal is not to hide complexity. The goal is to make it manageable.

Cost Truth means the app should know what actions cost, what was charged, what provider estimate was involved, and where there may be risk buffers, retries, or refunds.

This matters because OuiDire is not just a demo. It is a real tool built on real compute.

17. Technology stack and startup readiness

OuiDire is built on production-oriented infrastructure, not on a single AI prompt.

Each technical layer answers a trust question:

OuiDire is built on cloud and AI infrastructure designed for structured document audit.

Google Cloud, Gemini, and OpenAI GPT support distinct roles inside the workflow: extracting text, generating machine suggestions, producing deeper critical reads, tracking runs, observing costs, and preparing exports. OuiDire uses Gemini through Google Cloud's governed AI infrastructure by choice, including Gemini Enterprise Agent Platform, formerly Vertex AI, because the audit workflow involves sensitive records and should not rely on consumer-style data handling.

The result is a traceable review process: documents become cards, cards become findings, findings become digests and briefs, and each layer remains connected to the work that produced it.

18. Technical architecture

OuiDire is designed as a layered document-audit system.

A visible user action can trigger a chain of internal operations:

The point of this architecture is traceability.

OuiDire should not merely answer a question about a PDF. It should preserve how the answer was produced: which document was processed, which cards were created, which suggestions were machine-generated, which findings were human-confirmed, and which export synthesized the work.

19. Reliability and fallback design

OuiDire is built for imperfect documents and imperfect provider runs.

Psychiatric and legal records often arrive as scanned PDFs, image-based documents, forms, low-quality exports, or files with uneven formatting. A single extraction method may not always be enough.

Sometimes the user does not start with a PDF at all. They may have only a paper document received from a bailiff or another institution. In that case, the user can photograph the pages and use those images as the starting point for OCR, so the paper record can still become reviewable cards.

When OCR is required, OuiDire can use Google Cloud Document AI and is designed to support fallback routes when available. If one OCR path is unavailable, incomplete, or unsuitable for a given file, the system can try another route instead of treating the document as automatically unusable.

The same reliability principle applies to analytical runs. Large documents are not treated as one vague prompt. OuiDire separates work into scopes and chunks so provider runs can be tracked, retried, repaired, compared, and reconciled with cost data.

A user action may contain several internal operations. Full Analysis: Both Macro Sets, for example, is one visible action, but it may involve multiple provider runs, chunks, retries, or repairs. OuiDire's cost model therefore needs to distinguish what the user clicked, what the system actually ran, what the provider estimated, and what should finally be charged or refunded.

The goal is not to hide failure. The goal is to keep the workflow closer to a recoverable state:

OuiDire is not robust because it never fails. It is robust because it is designed to fail in recoverable ways.

20. What OuiDire is not

OuiDire is not a chatbot wrapped around a PDF.

It is not designed to produce a single answer from a whole file and ask the user to trust it.

It is not a replacement for legal, clinical, or expert judgment.

It is not a black box that hides the difference between source text, machine suggestions, human-confirmed findings, provider costs, weak extraction, and reviewed material.

OuiDire is designed to preserve those distinctions.

21. Privacy, control, and safeguards

OuiDire deals with sensitive records.

The app must therefore be designed around restraint, transparency, and user control.

Important principles:

The user remains responsible for review, selection, and interpretation.

OuiDire can help expose flaws in a record. It does not decide what a court, lawyer, clinician, or institution will accept.

The app's job is to make the file more readable, more structured, and more contestable.

22. What makes the invisible work important

Much of OuiDire's value is not visible at first glance.

The user sees cards, buttons, maps, and exports.

Behind that, the system has to handle:

This invisible work matters because it turns AI from a chatbox into an audit workflow.

OuiDire is not valuable because it has a Gemini button.

It is valuable because it structures the work around reviewable evidence, human judgment, traceability, and exportable findings.

23. Limits

OuiDire has limits.

OCR can be imperfect.

AI suggestions can be wrong, incomplete, overconfident, or irrelevant.

Weak input can produce weak output.

A Master Brief based on too little reviewed material should not be treated as a complete file analysis.

OuiDire does not provide legal advice, medical advice, diagnosis, or expert testimony.

It helps users organize, review, annotate, and explain records. Professional judgment remains professional judgment.

The safest way to use OuiDire is to treat it as an audit assistant:

24. Why this matters

Psychiatric records can shape people's lives.

They can affect liberty, treatment, credibility, litigation, family relationships, housing, employment, and access to justice.

When a record contains unsupported claims, recycled narratives, unclear sources, or distorted interpretations, the harm is not only informational. It can become institutional.

OuiDire exists because records should be answerable.

A file should not become final simply because it is long, formal, and difficult to read.

OuiDire helps turn dense records into structured review. It helps users see patterns, preserve objections, compare human and machine readings, and produce material that can be discussed with care.

The purpose is not to replace expertise.

The purpose is to make expertise easier to aim.

25. Core positioning statement

OuiDire is an audit workspace for psychiatric records.

It turns dense files into reviewable cards, separates source issues from narrative mechanisms, supports human annotation and machine suggestions, and produces exports that make the record easier to explain, challenge, and discuss.

The visible workspace is only the surface.

The real work is the layered audit system behind the interface: OCR, cards, source review, human decisions, machine suggestions, cost tracking, and exports working together to make dense records reviewable, traceable, and answerable.