Kshitij Gupta

Architecting a GenAI-powered KYC framework validating 5Cr+ customer records, combining fuzzy matching, multi-field address decomposition and cross-record comparison, and a prompt-engineered GenAI adjudication agent (tuned decoding parameters, structured prompting), improving onboarding match success by 10% and reducing hallucinations.
Developed a GenAI-powered dual vehicle ownership detection model identifying low-risk, high-profitability customer segments, by analysing co-insured vehicle patterns, financial behaviour, and historical usage data in partnership with underwriting and product teams, uncovering 10,000+ pricing and retention opportunities annually — with Phase 2 extending the model to new customer segmentation and targeted outreach.

Built a Customer Knowledge Repository consolidating policy, claims, financial behaviour, and demographic data from 5+ cross-team warehouses, by designing unified schemas in Python/SQL and automating data completeness assessments — reducing manual effort by 50% and improving KYC compliance accuracy by ~50%.
Delivered a Straight-Through Processing (STP) ML classification model for insurance claims, by engineering cross-domain features spanning financial, behavioural, demographic, and claims data with claims operations, achieving 75%+ AUC-ROC on validation — reducing manual adjudication by ~35% and cutting cycle time by ~24 hours.

Built a real-time Fintech Index Dashboard with an integrated GenAI chatbot, by aggregating data from 8+ financial technology sources and embedding a conversational LLM interface for natural language querying, liaising with international and in-house research stakeholders, enabling 15+ researchers to extract insights 40% faster and replacing 3 manual reporting workflows.
Provided technical consulting and GenAI integration support to 20+ researchers and PhD students, by resolving data pipeline issues within 24 hours, implementing research algorithms in Python, and augmenting academic models with LLM-based result interpretation — engaging directly with faculty supervisors — achieving a 90%+ satisfaction rate and boosting productivity by 30%.