Judgment to action plan
Artha Judge
A court-judgment workflow that turns PDFs or CCMS-style inputs into evidence-backed, human-reviewed action plans and a trusted dashboard.
Problem / itch
Court judgments often contain obligations, departments, deadlines, and directions buried inside dense PDFs. The itch was to convert that into reviewable action plans while keeping every claim attached to source proof.
What I built
- Built upload, mock/real-shaped CCMS intake, PDF profiling, OCR routing, deterministic extraction, optional LLM enrichment, and source-backed review packets.
- Extracted case metadata, parties, departments, disposition, legal phrases, directions, deadlines, action owners, priority, ambiguity flags, and escalation recommendations.
- Added human review states, evidence-required approval, approximate PDF highlights, duplicate detection, audit history, dashboard filters, CSV/JSON export, and grounded record Q&A.
- Added /judgments/evaluate so the same pipeline can be run over benchmark PDFs and compared against expected fields/actions.
Results
Turns a judgment PDF into case metadata, parties, departments, disposition, directions, deadlines, action owners, ambiguity flags, and source-backed action items.
Keeps AI output out of the trusted record until a human reviewer checks, edits, approves, rejects, or escalates it.
Attaches page-level evidence and highlighted PDF proof to every proposed field or action so reviewers can see why the system suggested it.
Includes a dashboard for approved action plans with status tracking, filters, audit history, CSV/JSON export, duplicate detection, and grounded Q&A over a record.
Routes clean digital PDFs through embedded text and messy scanned/corrupted PDFs through local vision OCR when needed.
Status
Currently halted; not actively working on it right now.