04

Parakh

Verifying Recruitment Offers for Nepali Migrant Workers: A Provenance-Aware, Bilingual Decision Support System

Published · 2026

Project Originator & Application Developer / Research Contributor
Public-interest technologyData provenanceInformation verificationOCRGovernment-data integration
Overview

Parakh originated from the idea that the information needed by Nepali migrant workers already exists across Department of Foreign Employment systems, but accessing and interpreting it can be fragmented and unintuitive. The project was originally built for the UNESCO Youth Hackathon. Parakh integrates information such as recruitment-agency records and lot/pre-permission information and presents it through a simpler verification workflow.

Research Question

Can existing Department of Foreign Employment information be reorganized into a provenance-aware, bilingual decision-support system that lets Nepali migrant workers verify recruitment offers more easily, without creating a new source of authority?

My Contribution
  • Proposed the original concept behind Parakh and helped define the initial problem and product direction.
  • Helped frame the project around making existing Department of Foreign Employment information more accessible and understandable rather than creating a new source of authority.
  • Built major parts of the application used for the project.
  • Worked on integrating the different verification flows into a unified interface where users can search information such as agency names and lot numbers.
  • Participated in the initial research framing and system architecture.
  • Contributed to the OCR evaluation work, including preparation and collection of advertisement-image data used for testing.
  • Later contributed to strengthening, restructuring, and reviewing the research manuscript.
Methodology
  • Collected and structured existing Department of Foreign Employment data, including recruitment-agency records and lot/pre-permission information.
  • Built a verification workflow that lets users search by agency name and lot number through a unified, bilingual interface.
  • Evaluated OCR on collected advertisement-image data used to surface recruitment offers.
  • Designed the system as provenance-aware, showing the source of each piece of information rather than presenting it as an authoritative verdict.
Key Findings
  • Existing official data could be reorganized into a simpler verification flow that is more navigable for migrant workers.
  • A provenance-aware presentation helped users interpret information without treating the system as a new source of authority.
  • OCR-based extraction of advertisement data showed how non-structured print offers could be surfaced for verification.
What I Learned
  • How to build public-interest technology around real human workflows.
  • How to integrate and present government data with provenance in mind.
  • How to incorporate OCR into a verification pipeline and evaluate it against noisy real-world images.
Limitations
  • The system depends on the completeness and quality of the underlying government data it surfaces.
  • OCR performance on real advertisement images is limited by image quality and formatting variability.
  • Rajan Raj Acharya took a larger role in the later research development and was therefore appropriately listed as first author.
Paper / Code / Artifacts