DEV3Technologies
All case studies

StartupXYZ

AI document verification that clears 85% of onboarding automatically

A Gemini vision pipeline extracts and cross-checks identity documents, scores authenticity in code and routes only uncertain cases to human reviewers.

GeminiPythonNode.jsMongoDBDistributed QueueAWS
Industry
Fintech
Region
India
Duration
10 weeks
Team
4 engineers
Services
AI Implementation

Impact

Applications needing manual review

15%85%

Before: 100%

Median verification time

4 min100%

Before: 1,560 min

Cost per verification

$0.2067%

Before: $0.60

Live project preview

Client systems are private — screenshots shown where permitted.

StartupXYZ — preview under NDA

Production screens available under NDA on request

The challenge

Every new customer’s identity documents were reviewed manually. Average verification took 26 hours, reviewers were the bottleneck for growth, and inconsistent decisions created compliance risk.

Our solution

Structured extraction

Documents are classified and fields extracted with Gemini using a strict JSON schema; every output is validated before use.

  • Name, date-of-birth and expiry checks are computed in code, not trusted from the model.
  • Scores above 85 auto-approve, 60–85 route to review, below 60 reject with reasons.
  • Reviewers see a split viewer with extracted fields and anomaly flags.

Evaluation harness

A labelled set of 2,000 documents gates every prompt or model change.

Results

  • Manual review dropped from 100% to 15% of applications.
  • Median verification time fell from 26 hours to 4 minutes.
  • Cost per verification reduced by two thirds.

Related case studies