Datics AI
← All work

Transcripts U.S.

Court reporting agencies publish transcript metadata without the file ever leaving their browser

Transcripts U.S. connects court reporting agencies with attorneys who need deposition transcripts. We changed its live marketplace so agencies select a transcript instead of uploading it, and title page details are extracted in their browser, OCR included.

Government, Legal & DefenseLegal transcript marketplaceAI added to a live product

Agencies review title page fields read in their browser before they publish or suppress a record.

The challenge

Agencies uploaded full transcripts, so the platform held confidential records it could read. The client wanted discovery through title page details only, with agencies in control.

What we did

We changed the trust model in stages: browser extraction for TXT, DOCX, PDF and scans, agency confirmation of every record, and one Request ID from inquiry to release.

The outcome

Every workflow item on the client's own checklist was delivered before the pilot, and agencies now publish title page details without the platform ever holding the file.

Client checklist
Every workflow item delivered before the pilot
Who confirms
The agency, field by field, before anything is published
Reads
TXT, DOCX, PDF and scanned pages, in the browser
Sent to the platform
Only the title page fields the agency approves

The starting point

Transcripts U.S. runs a national marketplace where court reporting agencies make deposition transcripts discoverable, and attorneys find, request, pay for and receive copies. The platform already had attorney, Stripe payment, fulfillment and delivery flows when we started.

What had to change

Agencies uploaded full transcripts, so the platform held confidential records it could read, and agencies wanted more control over access, pricing and release. The client moved to metadata only discovery, where attorneys search title page details and see no transcript before approval and payment.

What could not break

  • No full transcript is visible before agency authorization and payment
  • Search never shows which agency holds a transcript
  • Live requests, payment and delivery keep running on the platform
  • Extraction reads only the title page, never the body text
  • Attorneys and agencies never contact each other off the platform

The calls we made

We began with an actual versus required assessment of the live platform, then changed its trust model in stages instead of rebuilding it.

  1. 01

    Change the trust model in stages

    The first phase was a limited pilot: plain text transcripts and title page extraction only, on the attorney, Stripe payment, fulfillment and delivery flows that already worked. PDF, DOCX, OCR, ZIP files and integrations came after.

    Ruled out

    A rebuild of a marketplace whose request, payment and delivery flows already worked.

  2. 02

    Select, not upload. Extract in the agency's browser

    Extraction, including OCR for scanned pages, runs entirely in the agency's browser, so the file never reaches the platform's servers during discovery and only the fields the agency approves are sent.

    Ruled out

    The old upload workflow, which put every full transcript on the platform.

  3. 03

    The agency confirms every field

    OCR reports a confidence level per field and flags failures, and a record goes live only when the agency reviews, edits and publishes it. Suppress replaced Delete, so a record keeps its history.

  4. 04

    One Request ID from inquiry to release

    Questions, request, approval, price, payment, release and every timestamp hang off one platform ID, which also measures agency response, payment and fulfillment times.

How the system fits together

selects a filefields,confidenceapproved fieldsindexedsearchesrequestscheckoutaudit eventsanonymous resultsCourt reportingagenciesSelect, confirm,publishAttorneysSearch, request andpayBrowserextractionTXT, DOCX, PDF, OCRfor scansMarketplace webappAngular 16 onCloudFrontAgency reviewConfirm, edit,publish or suppressRequest lifecycleQuestions, requests,one Request…ServerlessbackendLambda, API Gateway,CognitoStripeCheckout at theagency's priceApprovedmetadataTitle page fields inRDS MySQL
PeopleBuilt by DaticsClient's existing systemOutside serviceData storeTranscript files are read in the agency's browser and never reach the platform's servers for extraction; only the fields an agency approves are indexed for search.

What we built

Request lifecycle

Requests only against published records, agency pricing set per request in place of a page based formula, Stripe checkout at that price and complete audit events, all under one Request ID.

Attorneys search published title page details and submit a request that carries one Request ID through review, price, payment and release.

Dashboards and onboarding

Email verification, mandatory onboarding with role based user types, agency, requester and admin dashboards, and notifications for agencies and attorneys.

The agency dashboard counts total, searchable, suppressed and needs review records and shows requests at each of seven stages.
03

Browser side metadata extraction

Title page extraction for TXT, DOCX, PDF and scanned documents runs in the agency's browser, with OCR confidence per field and failure reporting, and the privacy notice says only the first page is used.

04

Agency review and publishing

Agencies review, edit and publish or suppress each record, with Published, Needs Attention and Suppressed statuses and the page count shown, and attorneys search only published records without seeing the agency.

05

Questions through the platform

Attorneys ask about a transcript through structured options, the platform relays the question to the agency when needed and returns a neutral answer, so neither side sees the other's contact details.

How it works

Select

An agency selects a transcript on its own computer instead of uploading it.

Extract

The browser reads the title page, using OCR for scanned pages, and fills case number, date, state, witness and parties.

Confirm

The agency checks each field and publishes or suppresses the record.

Discover

Attorneys search published title page details without seeing which agency holds the transcript.

Request

Questions and requests run through the platform under one Request ID that tracks approval, price, payment and release.

A closer look

What changed

By the pilot, every workflow item on the client's own checklist was delivered, with only a cleanup of the demo environment left. The marketplace now indexes approved title page details without ever holding the transcript during discovery, and the platform moved into a controlled pilot with agencies and attorneys.

Every workflow item delivered
Client checklist
In a controlled pilot with agencies and attorneys
Status
“What impresses me the most is their belief in our vision and willingness to work through challenges alongside us.”
Keli McGiltonCo-Founder, Transcripts U.S.

Under the hood

Datics role
AI added to a live product
Data and security boundary
Extraction runs in the agency's browser and only approved metadata is sent; search hides the agency's identity, questions are relayed through the platform, and metadata has anti-scraping controls.
Infrastructure
Serverless AWS: Users, Transcripts, Purchases and Content services on Lambda behind API Gateway, with Cognito, RDS MySQL, SQS and S3 in AWS CDK, and the backend stack cut from 501 to 385 resources to fit CloudFormation's limit.
Integrations
Stripe checkout at the agency confirmed price, tied to the Request ID, Cognito sign in, and queued emails built from MJML templates.
Front end
Angular 16 with NgRx, AWS Amplify, Bootstrap 5 and Chart.js, hosted on CloudFront and deployed through CodeBuild.

Cloud and infrastructure

AWS LambdaAmazon CognitoAmazon RDS MySQLAWS CDK

Apps and APIs

Stripe

Also in the stack

Angular 16Browser OCRAPI Gateway

How it shipped

  1. Actual versus required

    We assessed the live platform against the client's revised workflow for metadata only discovery and proposed starting with a limited pilot.

  2. Pilot phase

    Plain text title page extraction, Request ID tracking, attorney requests, agency approval and pricing, admin visibility and role based access, on the existing payment and delivery flows.

  3. Select, not upload

    Browser extraction for DOCX, PDF and scans, suppress in place of delete, richer request statuses, audit events, dashboards and notifications, each story through QA.

  4. Pilot readiness

    Recorded attorney, agency and admin walkthroughs, then a round of the client's own requests, from dashboard counts and amber review notices to questions routed through the platform.

Built against their checklist

The client wrote its own checklist for metadata only discovery and asked for our recommendations on extraction, secure transfer, previews and wording, and we built against it item by item.

QA and recorded demos

Every story went through our QA with P0 to P2 bug lists, and recorded attorney, agency and admin walkthroughs let the client review each workflow before the pilot.

Plain answers for a non technical team

When the client asked whether agencies now upload twice, we walked through the flow in plain words: during discovery the file is only selected and read in the browser, never stored.

What this unlocked

Pilot feedback showed that title page details alone can reveal private information, so we proposed counsel of record matching and private, agency routed requests in place of open search, and a nightly sync from the court reporting software agencies already use.

Related work

Have a product with the same kind of constraints?

Tell us what already exists and what you want it to do. We will tell you what it takes.

Book a 30-minute Product Review