Project

Settlement & Advance Payment Automation Platform

From settlement creation to approval, AI automatically processes advance payment receipts

Service Type

AutomationWeb DevelopmentData

Tech Stack

Upstage Document AIOpenAI GPT-4o-mini

Key Features

Settlement Auto-Generation & Approval WorkflowAI OCR Receipt Auto-RecognitionAI Compliance VerificationQuote Parsing & Vendor Matching

Development Period

8–12 weeks

Settlement & Advance Payment Automation Platform

Problem

Still Processing Advance Payment Settlements Manually?

The daily reality for enterprise finance teams.
Manually entering receipts one by one, building Excel settlement reports, and cross-checking every supporting document.

Manual Receipt Entry

3–5 min per receipt

Excel Settlement Creation

30 min–1 hr per report

Manual Document Cross-Check

5–10 min per item

Manual Quote Entry

10–20 min per quote

Manual Receipt Entry

3–5 min per receipt

Excel Settlement Creation

30 min–1 hr per report

Manual Document Cross-Check

5–10 min per item

Manual Quote Entry

10–20 min per quote

3–5 min

Per receipt processing

1 hour

Per settlement report

30 min+

Document review time

50+

Policy review pages

Solution

Fromreceiptuploadtosettlementapproval—automatedbyAI

A B2B SaaS platform that automates the entire settlement process for enterprise finance teams. AI OCR automatically recognizes receipts from advance payment spending, and digitizes the full workflow from settlement creation to compliance review to final approval. What used to take hours of manual work now completes in minutes, while AI compliance checks prevent missing documentation and policy violations.

Product

Live Product Screenshots

Receipt management and detail view

Settlement detail and AI review screen

Data Flow

Receipt → Settlement Automation Pipeline

Upload

OCR

Normalize

Store

Settlement

Review

Compliance

Approve

Step 1

Receipt Upload

Upload receipts via drag & drop or camera capture → saved to Supabase Storage

Step 2

AI OCR Recognition

Upstage OCR extracts text. Dual mode support: standard OCR / high-precision Document Parse

Step 3

LLM Normalization + Classification

GPT converts to structured JSON + Upstage Classifier auto-categorizes expense types (parallel execution)

Step 4

DB Storage

Normalized data saved to receipts table. RLS enforces data isolation between projects

Step 5

Settlement Generation

Combines receipts + corporate card transactions to auto-generate settlement reports

Step 6

AI Compliance Review

Hard-coded rules + LLM semantic review detects policy violations, limit overruns, missing documentation

Step 7

Compliance Judgment

Issue classification (policy violation · limit exceeded · missing receipt · classification error · anomaly) + auto-assign 3-level severity (Low/Medium/High)

Step 8

Approval Workflow

Finance team review → Approve/Reject. Per-item comments & discussions + review snapshot audit trail

Receipt UploadUpload

Upload receipts via drag & drop or camera capture → saved to Supabase Storage

AI OCR RecognitionOCR

Upstage OCR extracts text. Dual mode support: standard OCR / high-precision Document Parse

LLM Normalization + ClassificationNormalize

GPT converts to structured JSON + Upstage Classifier auto-categorizes expense types (parallel execution)

DB StorageStore

Normalized data saved to receipts table. RLS enforces data isolation between projects

Settlement GenerationSettlement

Combines receipts + corporate card transactions to auto-generate settlement reports

AI Compliance ReviewReview

Hard-coded rules + LLM semantic review detects policy violations, limit overruns, missing documentation

Compliance JudgmentCompliance

Issue classification (policy violation · limit exceeded · missing receipt · classification error · anomaly) + auto-assign 3-level severity (Low/Medium/High)

Approval WorkflowApprove

Finance team review → Approve/Reject. Per-item comments & discussions + review snapshot audit trail

Features

Key Features

Advance Payment Receipt Processing

Drag & drop upload, multi-page grouping, AI OCR auto-recognition (merchant name, amount, date), automatic expense categorization

Automated Settlement Generation

Auto-generates settlement reports by combining receipts and corporate card transactions. Structured document format with per-settlement AI review session management

AI Compliance Review

Automated review based on evidence rules. Issue classification (policy violation · limit exceeded · missing receipt), 3-level severity judgment

Quote Parsing

Supports PDF, Excel, image, and DOCX. AI auto-extracts items, amounts, VAT, and terms; vendor matching by business registration number

3-Layer Architecture

Strict separation of Component → Hook → Service layers. 20 service functions × 17 TanStack Query hooks for consistent data access patterns

AI Cost Dashboard

Auto-logs all AI API calls, calculates USD costs. Per-project, per-feature, and per-user AI cost monitoring

Technical Strategy

Core Technical Strategies

Dual OCR Pipeline Modes

Receipt OCR offers parallel mode (3 API calls, ~$0.006/receipt) and combined mode (2 API calls, ~$0.002/receipt). High-value receipts prioritize accuracy; high-volume small receipts prioritize cost efficiency — customers choose the optimal processing mode for their usage patterns.

67% AI cost reduction$0.002–$0.006 per receipt

Hybrid AI Compliance Review

Combines hard-coded rules (amount limits, required evidence types) with LLM semantic review (contextual policy analysis). Quantitative rules are judged with 100% accuracy instantly; context-dependent rules such as 'taxi fares over $35 require justification' are supplemented by the LLM.

Hard-coded rulesLLM semantic review100% accuracy

Row Level Security Multi-Tenancy

Data isolation between projects is enforced at the PostgreSQL RLS level rather than the application level. Even if service code omits a WHERE project_id clause, data from other projects is never exposed — security breaches are prevented at the source.

PostgreSQL RLSData isolationSecurity at the source

SSE Streaming Real-Time Processing

Receipt OCR and quote parsing stream real-time processing progress to the client via Server-Sent Events. Even when batch-processing 100+ receipts, users can instantly see the status of each file.

SSE streamingReal-time progressBatch processing

Integrations

Integrated Services

Upstage Document OCR

Text extraction from receipt images/PDFs. High-precision recognition optimized for Korean receipts

Upstage Classification

Auto-categorizes receipt expenses (meals, transportation, office supplies)

Upstage Document Parsing

Parses quote PDFs into structured data (tables, line items, amounts)

OpenAI GPT

OCR text → structured JSON normalization, AI compliance review

Supabase Auth + Storage

Google/Microsoft OAuth SSO, file storage with 24-hour signed URLs

Resend

Transactional email delivery for approval request notifications and settlement completion alerts

Impact

Impact & Results

Receipt Entry

95%

Before

3–5 min per receipt

After

5–10 sec per receipt

Settlement Creation

90%

Before

30 min–1 hour

After

2–3 minutes

Document Review

98%

Before

5–10 min per item

After

3–5 sec per item

Quote Entry

95%

Before

10–20 min per quote

After

30 sec–1 min per quote

95%

Processing time reduction

$0.002

Cost per receipt

97%

OCR recognition accuracy

3-Layer

Architecture

Need a similar project? Get in touch with IMPAKERS.

Get a Free Consultation