B2B SaaS / Document AI
Evidra
Document processing workspace that turns invoices and scans into structured data, with field-level review, team comments, and approved exports.
Role & Contribution
Project creator · Product design · UI/UX design · Full-stack development
Built the product independently, from the initial idea and user flows to interface design, frontend, backend, and integrations.
Platform
Responsive web application with a multi-tenant API, background worker, and a Python document service.
Stack
Next.js, React, TypeScript, NestJS, Prisma, PostgreSQL, Python / FastAPI, OpenAI API, Docker
Interface Gallery
Ten views of the document workflow, captured from the application with English-language sample data.
Workspace overview
Document register
Document upload
Original document
Invoice field review
Line items and totals
Field comments
Result approval
Export history
Integrations and API
Overview
Evidra brings document intake, recognition, and human verification into one B2B workspace. Finance and operations teams can work through incoming invoices, inspect extracted values and line items, discuss individual fields, and approve the data before exporting it. The product is being refined for beta use.
Challenge
An extracted invoice is not automatically ready for accounting. Suppliers use different layouts, a single incorrect amount can affect the result, and the team needs to know which values were checked. The challenge was to make that verification process clear while keeping a dense, everyday workspace comfortable to use.
Solution
I built a workflow from file intake to approved export: a searchable document register, background processing, confidence indicators, editable invoice fields and line items, reviewer assignment, and comments attached to a document or a specific value. Approval, rejection, and reanalysis have separate actions and states.
Product Flow
A reusable flow map shows how users or data move through the product from entry to outcome.
Upload
Recognition
Extracted data
Team review
Approval
XLSX / CSV / JSON
Key Features
PDF and image upload with per-file settings
Document search, filters, and processing status
Editable invoice fields with confidence indicators
Line-item editing and totals validation
Reviewer assignment and field comments
Approval, rejection, reanalysis, and revision history
XLSX, CSV, and JSON exports
Workspace members, roles, and extraction templates
Architecture / Technical Overview
This diagram stays intentionally high level and avoids exposing sensitive implementation details.
Next.js workspace
NestJS API
Background worker / Python document engine
PostgreSQL / Object storage
Technical Decisions
- • Keep the web app, API, worker, and document engine in separate monorepo packages with shared contracts
- • Run document processing outside the request cycle and persist its progress, retry state, and extraction result
- • Store document data revisions separately from discussion so restoring values does not treat a comment as a data change
- • Require approved data for export and keep export history tied to the source documents
Challenges
- • Preserving unsaved edits while document state refreshes in the background
- • Keeping processing status separate from the human review decision
- • Making field comments discoverable and linking them to the correct document and value
- • Adapting a dense review workspace to tablet and mobile layouts
Result
Evidra demonstrates a connected document workflow with a working review interface, persisted edits and comments, and export controls. The current stage is beta stabilization. The screenshots use fictional English-language invoices in the running application; they do not represent customer data or recognition benchmarks.
Next project
Secure Connectivity Platform
Subscription-based connectivity platform with device provisioning, account access, server visibility, and support workflows.