All Case Studies & Notes
Multilingual NLP, FinTechRead Full System Blueprint

PayMe — 28-Language Google MuRIL Collections & DPD Risk Stratification

Deploying Google’s MuRIL multilingual transformer and Days-Past-Due (DPD) delinquency roll-rate models over serverless AWS Lambda, Twilio, and WhatsApp APIs (+37% repayment lift).

ROLE & OWNERSHIP
Data Scientist (at Markytics)
CASE STUDY AUTHOR
Rushikesh Gholap
TECHNOLOGY STACK
Google MuRIL NLPPython & Scikit-LearnDjangoAWS Lambda, EC2 & CloudWatchWhatsApp Business APITwilio
ENGINEERING COMPETENCIES
Multilingual Transformer NLPDPD Roll-Rate Risk ModelingEvent-Driven Messaging Automation

Consumer lending collections across multilingual cohorts suffered high delinquency roll-rates when outreach used manual dialers and English-only notifications, wasting agent capacity on self-curing accounts.

  • DPD Propensity Segmentation: Modeled borrower transition probabilities across Days Past Due (DPD) buckets to calibrate outreach timing and restructuring eligibility.
  • 28-Language MuRIL Intent Routing: Applied Google’s MuRIL transformer encoder to classify inbound borrower replies—including code-mixed regional scripts—across 28 languages.
  • Serverless Omnichannel Dispatch: Automated two-way WhatsApp Business and Twilio workflows on AWS Lambda paired with hierarchical Django supervisor dashboards.

Increased borrower repayment recovery rates by 37% while automating routine follow-ups and pre-approved credit cross-selling.

Share this case study:

More Engineering Case Studies

RGX ARCHITECTURE CONCIERGELOCAL SPEC INDEX
Search across 12 system specifications, BioNTech 3-hop GraphRAG, Temporal + MCP durable execution, 0.68s GPU Whisper voice AI, or 2019–2026 experience.