HyperVerge AI Lending (Assignment)

Millions of semi-urban and rural Indians remain outside the perimeter of institutional credit due to poor financial knowledge, lack of credit history, poor connectivity, and severe trust deficiencies between lenders and borrowers. Borrowers have no formal credit history, low digital literacy, and limited trust in lenders due to hidden fees and complex processes. Lenders cannot assess new-to-credit borrowers effectively, face high underwriting costs for small loans, and struggle with fraudulent or unverifiable documents. The result is a broken lending ecosystem that excludes the people who need credit most.

Ujjwal Gramin is a voice-first, multilingual mobile application that facilitates transparent, fair, and immediate access to credit for rural and semi-urban India. Two core AI features drive the product: a Conversational Loan Assistant that guides users through loan applications via speech or chat in their local language — eliminating literacy barriers — and Holistic AI Credit Intelligence that builds explainable credit scores from alternative data sources including mobile behaviour, bill payment history, transaction patterns, and community testimonials. The app is designed to run on low-end handsets on low-bandwidth networks, targeting a final APK size of 10–15 MB through aggressive SDK optimization and on-device processing of lightweight AI tasks.

I structured the problem using a dual-persona Value Proposition Canvas, separately mapping the jobs, pains, and gains of the rural borrower (farmer) and the institutional lender (bank). I then prioritized 6 key barriers — including no formal credit history, low digital literacy, fraud, low connectivity, mistrust of loan terms, and limited language options — using an Impact vs Feasibility matrix. This led to the recommendation of two LLM-powered features placed on a 2x2 prioritization matrix. For the Conversational Loan Assistant, I specified NLU model fine-tuning on loan data, speech-to-text and text-to-speech for Indian languages, and chatbot flows built on frameworks like Rasa. For Holistic AI Credit Intelligence, I specified ML models (XGBoost, Random Forest, Deep Learning) trained on alternative data, with LIME/SHAP for explainability. I also documented the technical trade-offs between on-device and server-side processing, and designed a full offline sync and retry/resume flow for low-connectivity resilience.

Delivered a complete product concept covering problem framing, dual-persona value proposition canvas, problem prioritization matrix, feature design for two LLM-powered capabilities, feature prioritization matrix, 15+ wireframe screens across the full user journey, technical architecture, SDK optimization strategy, on-device vs server-side trade-off analysis, and a KPI framework. The concept demonstrates how AI can bridge the credit gap for rural India while remaining accessible on entry-level hardware and low-bandwidth networks.