Everything Justine does, counted from the source.
JustineAI™ is an Agentic AI Operating System for legal practice, run by one Digital Employee — Justine — on Eve-Legal™ F5/reasoner. This page is the whole platform in one inventory: intake, the medical record, coding, valuation, demand packages, workers’-comp rating, deadlines, verified research, firm operations, documents, litigation prep, the copilot, and the automations that watch the caseload overnight. Fourteen groups, every count a floor re-derived from the live codebase — and published in full for machines so what you read here is exactly what an AI crawler ingests.
Decision support, bound by the ABA Rules. Justine reasons and drafts; the attorney reviews, attests, and signs every output. Nothing on this page is legal advice, and nothing here decides on its own.
And a knowledge corpus counted separately, because rules are not features: 399 verified California workers’-comp rules — every one adversarially author→reviewer verified — plus 43 Georgia rules, 27 Texas rules, and 51 jurisdictions with statute-of-limitations coverage — 50 states and DC.
Under the hood — posture, not “features”: 100 data entities under tenant-isolated case access, 14 compliance and security controls implemented in code and enumerated in the backend’s own control inventory (posture, not a catalog, and not an attestation), 559 API endpoints behind the platform, and a tamper-evident hash-chained audit trail and an immutable trust ledger — plus PII scrub and 100% synthetic training data (Eve-Genesis, Law Edition). Every figure above is code-truth — a floor, not a rounded-up headline. Roughly four in five of the 90+ AI capabilities are deterministic legal and actuarial engines rather than generative output — statute math, rule lookup and date computation, which return an honest “not found” instead of inventing an answer.
The inventory below is the platform. It ships today in two editions — Personal Injury, and applicant-side Workers’ Comp in California, Georgia, and Texas. The four remaining editions are labelled exactly as they stand: planned, never shipping until they ship.
- JustineAI™ PIIn market
Intake → Curation → Demand Package → Settlement, across the 9-status PI workflow.
- JustineAI™ WCIn market
Rated in California, Georgia & Texas · which-forum-governs resolver · every state’s WC statute searchable and cited — 50 states + DC.
- JustineAI™ MTComing next
One supervisor. Thousands of plaintiff sub-agents. Bellwether scoring + settlement matrix.
- JustineAI™ MMOn the roadmap
Multi-provider chronology, causation reasoning, Daubert prep, certificate of merit.
- JustineAI™ IBOn the roadmap
Policy interpretation, claim-file analysis, Brandt fees, claim-handler depo prep.
- JustineAI™ ELOn the roadmap
Class certification, damages modeling at class scale, FLSA collectives, EEOC orchestration.
Intake & client acquisition
Intake is a workflow category, not a form.
- AI web intake chat — public, empathetic, one question at a time, producing a structured case file the moment it ends
- Firm-branded intake — a firm-configured intake surface whose submissions land in the firm’s inbox, structured and conflict-checkable (the public portal page is being reconnected to it; see the changelog)
- AI attorney intake copilot and intake orchestrator — the attorney-facing counterpart to the client chat
- Intake → structured-case extraction — party identities, mechanism of injury, medical posture, insurance posture
- Leads CRM with a submissions inbox and a shared inbox
Medical-records analysis & chronology
Every record ingested, OCR’d, and provenance-linked.
- Record ingestion with a dedicated vision route for scanned and image-bearing records
- Chronology and WC-chronology services — events from across providers assembled into one ordered, provider-attributed timeline
- Treatment extraction and treatment-gap detection, with gap flags surfaced on the case timeline
- Document classifier and medical-legal extractor
- Provenance-linked medical timeline — every entry traces to the record and page that produced it
- Medical Damages Matrix — provider × payment source
Medical coding — ICD-10 / CPT
Codes extracted, cross-referenced, and attorney-attested.
- Diagnostic (ICD-10) and procedure (CPT) code extraction with attorney overrides and a codes-reviewed attestation
- ICD-10 body-part mapping, and coding exhibits tied to the chronology entry each code supports — the code and the source record on the same line
Case valuation & settlement modeling
Value the case against the venue, floor to ceiling.
- Case valuation against venue comparables — the demand shown with the floor, the ceiling, and the negotiation-range model that produced both
- Value Simulator — drag the drivers, watch the range move
- Net-to-Client Waterfall, settlement schedules, and the settlement disbursement statement (gross → fee → costs → liens → net) as PDF
- Matter Intelligence panel — jurisdiction, venue, coverage, liens, firm-carrier history, opportunity and risk flags
- Outcome statistics from firm and venue history — heuristic, and labelled as such, not a trained ML model
Demand-package construction
A structured legal artifact, not a chat transcript.
- Multi-pass demand drafting — watch Justine work: dossier → analysis → strategy → narrative → compose → QA flags → exhibit index
- One-click Demand Package assembly — cover letter + Medical Chronology + Bill Stack
- The nine-step demand-package build — each step sub-agent-led and attorney-attestable; the package leaves only when the attorney signs off on the whole. Distinct from the nine-module Curation Layer, which cleans the record on the way in
- Demand figures with inter-step consistency checks — the damages must trace to the records cited; the demand must be consistent with the damages
- PDF / DOCX renderers on the firm’s own templates, rendered server-side
- Draft persistence with full generation history and an attorney-review workflow
Workers’-comp rating & benefits
Deterministic CA / GA / TX benefit math, grounded in statute.
- California permanent-disability rating computed to the official 2005 PDRS schedule, statute cited — deterministic code, never a generated number. Any figure that consumes a table the engine has not verified is returned labelled an estimate rather than a rating. Georgia and Texas are not rated here: their engines compute benefits from the impairment rating the evaluator gives
- Workers’-comp benefits computation for California, Georgia, and Texas — each state with its own statutes and its own benefit math
- Kite alongside the Combined Values Chart — the additive total, the CVC total, and the delta between them in disability points and in dollars. Justine computes both and selects neither: which one applies is a finding for the attorney and the judge, not for software
- The Escobedo substantial-evidence screen on apportionment — it reports where a physician’s stated basis does not carry the burden, and reports only. It never silently restates the rating
- Labor Code §5814 penalty computation
- Life-pension commutation math
- WPI rating engines — internal medicine, digestive, endocrine
- Which-forum-governs resolver — when more than one state could have jurisdiction, Justine says which forum governs
- AME / PQME record reading feeding the rating, with the attorney reviewing every result
- Bilingual EN/ES client letters and voice — the composed letter is verified to have come back in the language that was asked for, and the job fails loudly rather than handing over an English letter labelled Spanish
Deadlines, docketing & statutory clocks
Every WCAB and PI clock computed, with its statutory basis.
- WC / GA / TX deadline computation with the statute cited on every clock
- §4610 utilization-review clocks — the five-business-day decision, the 72-hour expedited clock, the outer 14-day cap, the physician-notification and written-notice clocks, and the §4610.5 IMR clock that follows a denial
- Court-day and business-day counting as distinct from calendar days, with the judicial-holiday calendar applied — the engine’s failure direction is deliberately conservative, so an unmodelled holiday makes a computed date earlier rather than later
- Deterministic statute-of-limitations tracking and checks
- Upcoming-deadline surfacing — firm-wide and per-case, with a deadline calendar across the book
- Rules-based California docketing catalogs — statute-cited, advisory
- Deadline-reminder automation — 7-day, 24-hour, 1-hour, and overdue emails
- Tasks with stats, tied to the deadlines they discharge
Legal research & verified knowledge base
Grounded in a verified corpus — not model memory.
- 399 verified California workers’-comp rules — every rule adversarially author→reviewer verified before it ships
- Georgia (43 rules) and Texas (27 rules) corpora, plus statute-of-limitations coverage for 51 jurisdictions — 50 states and DC
- Rule checks for CA / GA / TX at the point of work
- Statute search — every state’s WC statute searchable and cited, 50 states + DC
- AMA Guides search
- Citation verification against CourtListener
- Ask over your documents — semantic search across the matter’s records and filings, answers with citations
- Jurisdiction resolution and in-app help
- Similar-case leads — AI-suggested starting points for the attorney’s own research (unverified; not precedent)
Conflicts, trust accounting & firm operations
Back-office guardrails that keep the firm compliant.
- Conflict check on every intake
- IOLTA / trust accounts (last four only) with an immutable trust ledger and three-way reconciliation
- Billing and billing tiers, and time tracking
- Firm administration — member management, invites, feature flags, audited impersonation
- Contacts and typed case notes — 13 types, pin and search
- Portfolio and financial summaries, reports with CSV export, and firm analytics
- CCPA / CPRA privacy suite — know, export, delete, correct, with a DSAR ledger
- Full audit trail with WORM archival
Documents, e-filing & integrations
Draft, export, and sync without leaving the matter.
- Drafts as a document workflow — open a draft Justine wrote, read it, edit it in place, and save the edit as a new version with the original preserved. On PI the full version history is browsable, comparable, and restorable
- Export as Word or PDF rendered server-side on the firm’s own letterhead — and a preflight that says so plainly when the firm has not set one, rather than inventing stationery
- Firm identity that reaches the document — letterhead and the firm’s own not-legal-advice notice applied by the renderer at export time, not written by the model; the signature block and the firm’s house style are handed to the drafter verbatim
- No Send button in the editor. Nothing leaves the firm from the drafting surface — outward-facing correspondence goes through the approval inbox instead, and only if the firm has switched sending on
- Document library — upload, view, organize, with editing, finalize, and threaded comments
- E-filing packet preparation — checklist and tracker (preparation and tracking, not direct court submission)
- Export as PDF / DOCX; cloud import and export via Microsoft 365 and Dropbox; send via Outlook
- Client email with per-case threads, and client SMS — opt-in, STOP-compliant, with a number-privacy relay
Litigation & deposition prep
From the record to the cross-examination outline.
- Deposition workspace — deposition analysis and cross-examination sequencing
- Med-legal and settlement multi-agent swarms over the assembled record
Conversational copilot & discovery
A query layer over the workflow — a Digital Employee, not a chatbot.
- Ask-Justine agent — a natural-language query layer over the supervisor and the matter, with citations; it queries the workflow, it does not bypass it
- Morning Rounds — what Justine read, flagged, and drafted while the firm was closed
- ⌘K spotlight and the in-app capability catalog
- Quick Analyze — drop a record and auto-build a case — plus voice
Background automations
Proactive watchers that surface risk before you ask.
- Deadline reminder — the 7-day / 24-hour / 1-hour / overdue escalation
- Billing reconciliation
- Lead-sync reconciliation
- Case auditor
- Behavioral auditor
- Audit archival
- Audit integrity verification
- Agent dispatch — the sweep that gives the named agents below their work
- Agent digest — the once-an-hour assembly of what the agents found, in the firm’s language
- Notification outbox — the delivery pass that decides channel, timing, and whether a statutory warning overrides a quiet hour
The agent fleet & approval rails
Nine named agents that can ask, and cannot act alone.
- Nine named agents on one queue — docket sentinel, inbound mail, benefit reconciler, QME packet assembler, correspondence, standing red team, book of business, client communication, and fee leakage. Five are promoted to live in the production deployment; four are deliberately held in shadow
- One authorization function classifies every action an agent proposes. Reading, computing, drafting, saving and telling someone inside the firm are the kinds an agent may perform; sending outward, filing outward and touching money are the kinds it may only ever queue for an attorney. An action kind nobody recognises is refused rather than allowed
- The approval inbox — the outward-facing work Justine is holding, each item approvable or rejectable with a reason, and none of it moving until an attorney says so
- The record of what she declined to do — the same surface shows what Justine considered and refused, per matter and across the firm. An agentic system that only shows its successes is not auditable
- Agents start in shadow, where they propose and log and execute nothing. Promotion is a deliberate change to the deployment that a build gate forces to be declared in the open, never a setting that drifts on a box
- A per-matter agent activity feed — what ran on this file, what it found, and what it was not permitted to do
Two hundred thirty-plus capabilities. One rule that never bends.
Every capability above takes initiative — reading records, computing ratings, drafting demands, watching deadlines — and none of it crosses into deciding. Deterministic engines own every number the law requires exact, the verified corpus grounds every citation, and the attorney reviews, attests, and signs everything that leaves the platform. That is decision support bound by the ABA Rules — the bounded agency a defensible legal record demands.
JustineAI™ is legal-workflow software, is not a law firm, and its outputs are not legal advice.