Language as the Greatest Digital Divide
In India, English is spoken by less than 12% of the population, yet more than 60% of official government websites use English as their primary interface. Even when Hindi translations exist, they are often literal, stiff translations of bureaucratic terminology that everyday citizens do not use.
Handling Code-Mixed "Hinglish" Queries
Citizens rarely search using pure formal Hindi ("उपायुक्त का कार्यालय कहाँ स्थित है?") or formal English ("Where is the office of the Deputy Commissioner?"). Instead, they ask in code-mixed colloquial vernacular:
"Karnal me Kisan credit card ke liye kaunse documents chahiye aur kiske paas jana padega?"
Our NLP & Tokenization Strategy
To process these complex queries accurately, City Mitra implements a multi-step semantic parsing pipeline:
- Intent & Entity Extraction: Detecting the core civic intent (e.g. Scheme_Requirements, Officer_Lookup, Emergency_Helpline).
- Location Disambiguation: Identifying implicit district context from the user's active session or explicit mention in the prompt.
- Phonetic Transliteration Normalization: Mapping romanized Hindi variations (e.g., "yojna", "yojana", "scheem") to the standardized knowledge graph entity.
Zero-Friction Access for All Citizens
Our philosophy is simple: technology should adapt to how people speak, not force people to learn how computers think. By making City Mitra bilingual and conversational, we ensure that every citizen, regardless of tech literacy, can access public services with dignity and ease.