Context is scattered
PDFs, DOCX files, proof copies, notes, and reports are separated across tools and folders.
Agentic review, explicit human decisions.
Acribica
An agentic AI system for document-heavy review workflows.
A local review workspace for proofreading, page proofs, and editorial work—keeping source evidence, AI findings, human decisions, and export artifacts in one inspectable flow.
02The problem
Document review is still too fragmented.
Source files, proof copies, notes, findings, and reports often live beside one another without a shared decision trail.
Findings lose intent
Useful review comments need evidence, confidence, status, and a clear handoff path.
Extraction is imperfect
PDF text layers and export artifacts can create noise, ambiguity, and false positives.
03Product idea
Specialized AI roles turn documents into reviewable findings.
Acribica keeps document context, role-based findings, review decisions, and exports in one focused local workflow.
Evidence stays visible
Every finding remains anchored to the page, text, or document context that produced it.
Roles stay focused
Dedicated reviewers examine language, subject matter, methodology, formalities, and consistency.
Decisions stay explicit
Human reviewers confirm, reject, defer, or turn a finding into an author question.
04Workflow
From document to agentic review and human decision.
The pipeline separates extraction, specialized AI review, and human judgment so each stage remains legible and inspectable.
-
01
Upload
Load a PDF or DOCX as a local project.
-
02
Extraction
Read text, layout, structure, and quality signals.
-
03
Agentic role review
Run focused AI roles against extracted evidence.
-
04
Human review
Confirm, reject, defer, or create an author question.
-
05
Export
Generate reports and handoff-ready artifacts.
EvidencePDF · DOCX
FindingsContext · rationale
AI proposesspecialized roles
Humans decideexplicit outcomes
ArtifactsReports · data · proofs
05Architecture
Technical implementation: local, modular, inspectable.
Acribica is composed as a small set of clear layers rather than a single opaque automation.
- 01
React/Vite frontend
Split-view review UI with queue, detail panel, status controls, and export links.
- 02
FastAPI backend
Uploads, project state, analysis jobs, findings, and report endpoints.
- 03
Document pipeline
PDF/DOCX handling with text and layout extraction, including text-layer quality checks.
- 04
Role model
Specialized, high-capability AI agents for language, subject review, methodology, formalities, and consistency.
- 05
Local artifacts
Project data and exports stay in the local workspace.
06Review interface
AI proposes. Humans decide.
The interface is organized around evidence: where the finding appears, why it matters, and what the reviewer decides.
Document or page context
Highlighted finding
Findings queue
Rationale and suggestion
Explicit decision controls
07Outputs
Findings become usable handoff artifacts.
The review does not end inside the interface. Accepted decisions leave the workspace in formats suited to people and downstream systems.
Markdown report
Readable review summary for human handoff.
CSV / JSON
Structured findings for processing and integration.
Annotated PDF
Editorial context placed back onto the proof.
Correction list
Actionable items for production workflows.
08Project status
Current status and next steps.
MVP complete
The current build will soon be available on GitHub.
Contributions welcome
Pull requests and focused improvements are welcome.
Next refinement
Tighten tests, fixtures, documentation, and the public-ready project structure.
github.com/acribica
Developed with Codex.
Acribica makes document review more structured, inspectable, and handoff-ready.
Back to top