Accepting SMB clients now
·
Sprints from $1,200
·
Diagnosis in 24 hours
·
Available now

sample_diagnosis · sample · workflow_automation

Law Firm Document Classifier

tier_02 · prototype · 7 days from input to prototype

SAMPLE DIAGNOSISThis is an illustrative scenario, not a real client engagement. The methodology, reasoning, and build decisions are genuine. The business and metrics are fictional.

// the input

What they sent us.

verbatim · initial submission

We spend two days before every matter opening just sorting and labeling documents clients send in.

// the diagnosis

What we actually found.

  • New-matter intake consuming ~16 hours per week across two paralegals
  • Documents arrive as unsorted email attachments — no naming convention, no intake structure
  • 80% of documents fit 6 repeatable categories: contracts, correspondence, court docs, IDs, financials, other

the real problem

Not a volume problem — a classification problem. Documents arrive in bulk by email with no structure. Paralegals were manually reading every attachment before legal work could start. 80% of documents fell into 6 standard categories.

// what we cut

Everything we deliberately didn't build.

cut itemreason
Full document management system replacementExisting system works fine — the intake is the bottleneck, not storage.
Client portalChange management too heavy for a hypothesis test. Add in phase 2 if classification holds.
OCR for handwritten documentsEdge case in this practice. Marginal return doesn't justify the build cost in v1.

// what we built

The actual thing.

An email-intake processor that reads incoming attachments, classifies each document into one of six categories using Claude, and delivers a structured summary with suggested matter labels to the paralegal's inbox — ready for review, not from scratch.

PythonClaude APIGmail APIGoogle DriveRailway

// the outcome

What changed.

16 hrs → 3 hrs
per week on document intake
6 categories
auto-classified per matter
7 days
problem to working prototype

// start your project

Your bottleneck could be next.

Start a project →← All work