Document workflows / Local AI
Leng
Developed at Nile University
From scattered submissions to a report ready for review.
A desktop workbench for preparing institutional annual reports, with local AI extraction, source references, and quality checks in one place.
- Context
- Nile University / Institutional reporting
- Scope
- Application design and development
- Platform
- Desktop / Windows
- Status
- Public source available
From source material to a structured report. Editorial illustration for Leng.
The context
A report assembled from many sources.
Annual reporting brings together spreadsheets, PDFs, and written submissions from multiple owners. Figures need to agree, gaps need an owner, and reviewers need to trace statements back to the material behind them.
What we built
One workbench for report preparation.
Leng turns that preparation process into a single desktop workflow: bring in source files, organise report variables, review the collection matrix, resolve quality issues, and export a draft for human review.
Source-backed extraction
PDF, spreadsheet, and document inputs feed extraction proposals that retain file references, page context, and confidence information.
A structured report model
Sections and typed variables organise the report. Carry-forward values and owner-level checks help reviewers identify what still needs attention.
Review before publication
Quality checks surface missing values, date references, inconsistencies, and carried-forward content. Drafts and review packages can be exported for the next stage.
Design decisions
The choices behind the experience.
Run document analysis locally
Electron provides the desktop workspace, while llama.cpp and Qwen support local inference. OCR and document parsing run locally as well.
Retain the evidence
Accepted values keep their provenance, and import decisions form an audit trail instead of disappearing into a finished document.
Keep the reviewer in charge
Confidence and quality checks guide review. Generated drafts remain material for human approval, with manual corrections supported throughout.
The result
A traceable path from source to draft.
A working desktop application that connects document intake, structured preparation, quality review, and multi-format export. Reviewers can follow a value back to its source and carry an approved draft into the reporting process.
- Electron
- JavaScript
- llama.cpp
- Qwen
- Local OCR
- Document parsing
Inside Leng
Evidence stays close to the work.
