DocStream
Document Processing
DocStream is a document platform for invoice processing, designed around the field it reads badly rather than the fields it reads well. Confidence is shown, doubt is routed to a person, and the ledger is guarded from both.
Designed & developed by Entrovix
- Year
- 2024
- Discipline
- Document AIWorkflow DesignEngineering
- Services
- Workflow DesignUX/UI DesignBack-end EngineeringAI Integration
- Built with
- PythonCeleryOCR APILLM APIPostgreSQL
The Challenge
Wrong in a way nothing catches
Automated extraction fails politely. A misread amount is not flagged by anything downstream — it sits in the ledger looking exactly like a correct one, and the tool that produced it offered nobody a reason to doubt it.
So the pitch had to invert. Rather than promising a machine that reads everything, the product promises to know what it has not read well — and to make that moment of doubt cheap enough that a person settles it on the spot instead of re-keying the document.
The Approach
How we worked
Confidence is the interface
Every extracted field carries its own confidence, worn as a chip next to the value. Certainty is a property of each answer, never a mood the whole product projects.
Doubt goes to a person
A field below threshold is routed to review rather than passed along, and it arrives with the exact region of the source document already highlighted.
Make the check cost seconds
The review view puts the highlighted source and the extracted value side by side. Confirming or correcting one field is a glance and a keystroke, not an investigation.
Guard the ledger
An amount reaches the accounting system by one of two routes: high confidence, or a person's eyes. There is no third door.
The Solution
A pipeline that knows when to ask
Invoices move through a queue that states their condition plainly — processed documents alongside the amber one that needs review. Opening it shows the page itself, with extraction boxes drawn over the exact text each value came from.
Beside the preview, the extracted fields — vendor, invoice number, amount, due date — each wear their own confidence, and the doubtful one is flagged for checking. A person confirms it where it stands, and the document moves on.
Inside the product
What it actually does
Every Screen
One product, whatever it is opened on
The phone layout is not the desktop one squeezed down. Density, hit areas and reading order are decided per screen, and the interface keeps its meaning at both ends.
Capability
What we brought to DocStream
The same services are available on their own — each one links to what it involves, how it runs and what you get.
Built with
- Python
- Celery
- OCR API
- LLM API
- PostgreSQL
Disciplines
Document AI · Workflow Design · Engineering
The Outcome
What it settled
- Every field carries its own confidence, not a shared score
- Doubtful values reach a person with their source already highlighted
- A flagged field is settled with a glance, not an investigation
- Nothing posts to accounting without high confidence or human eyes
That was some of our work
Now let’s talk about yours.
Tell us what you’re building. You’ll get a written plan, a clear proposal, and a team that shows its work every week.