Back to systems
CASE FILE / AF-09REVIEWED EVIDENCEACTIVE SYSTEM / June 2026

AI legal operations / Human review boundary

PetitionReady

Built an AI-assisted bankruptcy intake readiness console for paralegals, focused on case blockers, document gaps, readiness scoring, client follow-up, and attorney-review packet preparation without crossing into legal advice.

Operating proof
Direct product and implementation evidence for an AI-assisted legal-operations workflow, spanning typed bankruptcy intake models, deterministic readiness analysis, server-side structured model outputs, fallback behavior, and explicit paralegal-to-attorney review boundaries.
Engagement
Personal Project
Evidence set
5 reviewed artifacts
Capability coverage
4 documented areas
Legal Operations UXAI Product SystemsFull-Stack EngineeringHuman Review Design

Primary artifact

01 / 05
artifact_viewer.sh

Matter readiness operating console

The working paralegal dashboard connects queue state, readiness scoring, blockers, next action, attorney-review items, and source-grounded Copilot evidence without hiding the human review boundary.

01 / Context

Situation

Exceptions surfaced too late

Bankruptcy firms lose time when intake packets are incomplete, income data conflicts, creditor addresses are missing, documents are rejected, or payment status is unclear. Those exceptions often surface late, after a paralegal has already started assembling a packet for attorney review.

The handoff lacked an operating view

The product needed to answer one practical workflow question: can this client matter move toward attorney review, and if not, what exact blocker should the paralegal resolve next?

Legal scope required a hard boundary

The prototype intentionally focuses on readiness, blockers, evidence, and attorney-review preparation instead of automated filing, legal advice, eligibility decisions, or petition generation.

02 / Mandate

Mandate

Make readiness inspectable

Design and build a working dashboard that gives paralegals a clear view of active matters, blocker counts, readiness scores, queue status, and next actions.

Ground AI in the current packet

Use structured AI outputs to summarize blockers, find conflicts, draft client follow-ups, and compile attorney-review briefs from the current case packet.

Preserve reviewer access without credentials

Make the demo useful without credentials through deterministic local fallback analysis, while preserving the same UI shape when model calls are unavailable.

evidence_task.log

Case-grounded Copilot evidence

The Copilot panel exposes its evidence snapshot alongside bounded modes for blocker analysis, conflict checks, client follow-up drafts, and attorney-review brief compilation.

03 / Build

Build

1.Modeled repeatable case states

Created seeded Chapter 7 and Chapter 13 matters with client facts, document states, retainer status, issues, readiness stages, raw notes, and blocker metadata.

2.Built the paralegal operating console

Implemented readiness summary cards, matter queue pagination, status badges, selected matter details, top blockers, next paralegal action, and attorney-review item surfaces. Readiness starts at 100, subtracts documented issue weights, and maps each matter to attorney review, cleanup, or blocked state.

3.Implemented structured server-side Copilot

Built server-side Paralegal Copilot modes for blocker analysis, conflict checks, client follow-up drafts, and attorney-review brief compilation using the OpenAI Responses API, structured JSON output, and Zod validation.

4.Kept deterministic fallback equivalent

Created local analysis for readiness scoring, blocker detection, follow-up generation, and brief assembly so reviewers can evaluate the full workflow without an OpenAI key.

evidence_action.log

Exception and notification queue

The notification state surfaces new matters, document exceptions, top blockers, and the next paralegal action without separating alerts from the operating queue.

Blocker-to-next-action inspector

The inspector keeps source notes, unresolved evidence, the next paralegal action, and attorney-review preparation in one inspectable matter state.

04 / Outcomes

Outcomes

One blocker-to-review workflow

The prototype turns intake state into a repeatable queue, readiness score, issue list, evidence snapshot, next action, and attorney-review handoff instead of scattering those decisions across disconnected views.

Inspectable human-in-the-loop boundary

PetitionReady grounds Copilot outputs in supplied case evidence, keeps paralegal and attorney review visible, and explicitly avoids legal conclusions, filing decisions, and automated petition generation.

Production-minded prototype with explicit gaps

The public repo and Vercel demo include typed domain models, structured AI response validation, focused Vitest coverage, timeout and fallback behavior, and seeded local data. Authentication, persistence, audit logs, and production legal workflows remain deliberately out of scope.

evidence_result.log

Selected-matter readiness review

A selected Chapter 13 matter connects its readiness score and payment-plan blocker to the evidence and cleanup action required before attorney review.

Continue the evidence trail

From proof to role fit

Compare PetitionReady with adjacent systems, or carry its reviewed capabilities into an Adaptive Focus brief.

Operational UXHuman-in-the-loop AIAI product systemsInternal tools