AutoFlow Studio is integrated into NexusVision as the business workflow automation module. It brings AI assistant, document processing, workflow automation, and business operations into the same Python desktop platform as NexusVision Lab and HoloCanvas Studio.
Many people still repeat office tasks manually:
Open PDF
|
Read
|
Summarize
|
Write report
|
Save DOCX
|
Send emailAutoFlow automates that pattern:
Upload
|
Extract
|
Analyze
|
Generate
|
ExportThe platform now covers three practical domains:
- computer vision experiments through NexusVision Lab
- gesture creative work through HoloCanvas Studio
- real business document automation through AutoFlow Studio
AutoFlow is especially relevant for recurring work like editing reports, preparing presentations, processing PDFs, creating school or university documents, and organizing project output.
Implemented in app/autoflow/:
- PDF/TXT/MD input
- PyMuPDF PDF text extraction
- grounded deterministic local summary draft
- source evidence snippets
- missing-information detection
- document intelligence table
- quality review checks
- template-specific output for reports, invoices, and study notes
- confidence level and reason
- workflow templates
- TXT export
- DOCX export
- SQLite workflow history
- PyQt6 dashboard panel
Current templates:
- Report Automation
- Invoice Intelligence
- Thesis Study Assistant
Input:
Laporan.pdf
Pipeline:
PDF
|
Text Extraction
|
AI Summary Draft
|
Generate DOCX
|
Generate TXT
|
ExportInput:
Invoice Supplier.pdf
Pipeline:
Invoice PDF
|
Extract Text
|
Summarize Supplier and Totals
|
Save History
|
Generate ReportInput:
Skripsi.pdf
Pipeline:
PDF
|
AI Reading Draft
|
Chapter Summary
|
Key Points
|
Flashcard-ready NotesAutoFlow should feel like a professional workflow tool, not a plain upload form.
In the desktop app, opening AutoFlow switches NexusVision into a focused document workspace. The webcam preview and camera metrics are hidden so the full central area is used for workflow controls, selected document status, and analysis output.
The local summarizer is intentionally conservative. It only uses extracted source text, includes evidence snippets, and marks missing values as not detected instead of guessing. This is especially important for invoices, where amounts, dates, and supplier contact details must be verified against the original file before approval.
Current output sections:
- Grounding Note
- Document Intelligence
- Executive Summary
- Key Points
- Source Evidence
- Detected Signals
- Missing Or Unclear Information
- Quality Review
- Template-Specific Output
- Recommended Actions
- Export Notes
- Confidence
Visual builder concept for a future phase:
[Upload]
|
[OCR]
|
[AI Summary]
|
[Generate DOCX]
|
[Export]Expected future workspace areas:
- template and node library
- visual workflow canvas
- node settings inspector
- run logs
- export center
- workflow history
Status: implemented in desktop form.
- Upload/select PDF or text file
- Extract text
- Generate summary draft
- Generate DOCX
- Generate TXT
- History
- Dashboard panel
- Workflow builder
- Drag and drop nodes
- Template system
- Automation rules
- Preview output per step
- Saved workflows
- Email automation
- Spreadsheet automation
- Batch processing
- AI agents
- Invoice field extraction
- Multi-file projects
PDF
|
AI
|
DOCX
|
PPTX
|
Email- Multi-step pipelines
- Human approval gates
- Scheduled workflows
- Team-ready workflow library
- Exportable audit trail
Current desktop runtime:
- Python
- PyQt6
- SQLite
- PyMuPDF
- python-docx
- Pydantic-ready contracts
Future Python web/runtime options:
- FastAPI
- Jinja2
- HTMX
- SQLAlchemy
- pdfplumber
- openpyxl
- reportlab
- OpenAI API
- Gemini API
- local LLM adapter
Hard boundary remains: Python-only, no React, no Next.js, no TypeScript, and no Node.js frontend frameworks.