Vernacular NLP: Engineering Hindi-English Code-Mixed Speech & Public Gazette Parsing
How NorAI built the Smart Dainik News ingestion pipeline to parse complex Indian public employment gazettes, Hindi PDF tables, and code-mixed vernacular announcements.
Dhruw Singh
Infrastructure & Reliability Lead
The Reality of Regional Indian Document Parsing
Public employment notifications (Sarkari Gazettes) in North India present unique document challenges: scanned physical printouts with ink smudges, complex multi-script Hindi/English (Hinglish) code-mixing, and critical eligibility tables embedded in unstructured PDF layouts.
Standard international OCR models fail significantly on Devanagari script conjuncts (युग्माक्षर) and misinterpret eligibility age criteria, causing thousands of aspiring job seekers to miss critical application deadlines.
Bilingual Entity Normalization Engine
Our pipeline extracts four core verified data points from every employment notification: Eligibility Qualifications, Age Limits with category relaxations, Important Application Deadlines, and Official Direct Submission Links.
By utilizing custom fine-tuned Devanagari OCR models paired with regex boundary detectors, we achieve 99.2% extraction accuracy across regional publications and local gazettes.
export interface GazetteJobDigest {
notificationId: string;
departmentName: string; // e.g., "UPSSSC / UP Police"
postTitleHindi: string;
postTitleEnglish: string;
totalVacancies: number;
eligibility: {
minEducation: string;
ageMin: number;
ageMax: number;
categoryRelaxations: Record<string, number>;
};
deadlines: {
applicationStart: string;
applicationEnd: string;
examDateTentative?: string;
};
officialLink: string;
verifiedStamp: boolean;
}Smart Dainik News & Job Digest
Verified public employment alerts and regional policy summaries for North India.