Skip to content
CraftSoftware System

AI & automation · Professional services

Triage AssistConcept

An AI intake flow that reads incoming documents, extracts key fields, and routes each case to the right reviewer for confirmation.

Concept interface with demo data. Not a client project.
Client
No client (concept)

Concept project — no real client. Created to show how we would approach this kind of product.

Industry
Professional services
Platforms
Web, Email
Delivery scope
  • Workflow analysis
  • Pilot with evaluation set
  • Review interface
  • Integration
  • Monitoring
Technologies
  • Python
  • LLM APIs
  • Queue workers
  • PostgreSQL

Overview

The scenario

Triage Assist imagines a firm that receives hundreds of documents a week by email. Staff currently open each one, retype the key details, and forward it to a colleague. The concept automates the reading and routing while keeping a person responsible for every decision.

The problem

Intake takes hours each day, urgent matters wait behind routine ones, and details are occasionally mistyped. The firm needs speed without handing judgement to a machine.

Our solution

Documents are classified and key fields extracted with a confidence score. High-confidence results are pre-filled for a quick check; low-confidence fields are highlighted for review. Every case is routed to a team queue based on type and urgency.

User journey

How someone actually uses it

  1. 1

    Document arrives

    An email with attachments lands in the intake mailbox.

  2. 2

    Read and classify

    The system identifies the document type and extracts fields with confidence scores.

  3. 3

    Route

    The case appears in the right team's queue, urgent items first.

  4. 4

    Review

    A reviewer confirms or corrects highlighted fields in one screen.

  5. 5

    Learn

    Corrections feed the evaluation set used to measure accuracy over time.

Web

Product screens

4 screens from the web application.

Screen 1 of 4: Intake queue sorted by urgency, with confidence per document.

Swipe, use the arrow keys, or pick a thumbnail. Screens are concept designs with demo data.

Read descriptions of all 4 screens
  1. 1. Web

    Intake queue sorted by urgency, with confidence per document.

    Intake queue listing documents with type, urgency, and confidence

  2. 2. Web

    Reviewers confirm fields side by side with the document.

    Review screen with document text on the left and extracted fields on the right

  3. 3. Web

    Routing rules are readable and editable by the team.

    Routing rules page listing conditions and destination teams

  4. 4. Web

    Quality dashboard tracks accuracy against the evaluation set.

    Quality dashboard showing volume and correction rate trends

Key features

What it does

  • 01

    Classification

    Identifies document types agreed with the team during the pilot.

  • 02

    Field extraction

    Pulls names, dates, references, and amounts into structured data.

  • 03

    Confidence highlighting

    Uncertain values are flagged instead of silently accepted.

  • 04

    Smart routing

    Rules send each case to the right queue by type and urgency.

  • 05

    Review queue

    Side-by-side document and fields for fast confirmation.

  • 06

    Quality dashboard

    Accuracy, volume, and correction rates tracked over time.

Approach

Design and development

Design

The review screen was designed around the reviewer's eye path: document on the left, fields on the right, uncertain values in amber, and keyboard shortcuts to confirm.

Development

Python workers process a queue of documents, call language-model APIs with structured output, and store results with the prompt version used, so every result can be audited.

More work

All projects
ConceptWeb platform· Distribution

Ledgerly Ops

An operations dashboard that unifies orders, stock, and invoices so a growing distributor can stop reconciling spreadsheets.

  • Next.js
  • TypeScript
  • PostgreSQL
View Case Study
ConceptMobile app· Healthcare

Clinic Queue

Appointment booking and live queue updates for outpatient clinics, so patients wait at home rather than in the corridor.

  • React Native
  • Node.js
  • PostgreSQL
View Case Study
ConceptWeb platform· Retail

Harvest Market

A fresh-produce storefront with delivery slots, substitutions, and an admin that the store team updates themselves.

  • Next.js
  • Headless CMS
  • Paystack
View Case Study

Want something like Triage Assist?
Let's build yours.

We'll start from your users and constraints, not a template. Tell us about your project and we'll reply with questions and next steps.