SpashtaAI

Inspiration

Over 300 million Indians remain excluded from formal financial services — not because they lack access, but because they can't understand. A farmer in rural Bihar can receive a loan document translated perfectly into Hindi, yet still not comprehend what "12% APR with compound interest" actually means for his family.

We realized: translation is not the barrier. Comprehension is.

What it does

SpashtaAI performs concept translation, not language translation. It transforms complex financial documents into explanations using familiar rural analogies:

Before: "Your loan accrues monthly interest at 12% APR with compound interest."

After: "You borrow Rs. 10,000. After one year, you pay back Rs. 11,200. The extra Rs. 1,200 is the fee for borrowing — like unpicked crops rotting and losing value each day you delay."

The system:

Detects financial jargon using NER models
Extracts the underlying concept using LLMs
Maps it to rural analogies (farming cycles, daily wages, local markets)
Personalizes explanations (converts percentages to actual rupee amounts)
Outputs in text + voice for low-literacy users

How we will build it

LLM: Phi-3 / Gemma 2B (small models for offline/on-device inference)
NLP: IndicBERT for Indian language understanding
Speech: Whisper for voice input, TTS for audio output
Frontend: React Native mobile app
Backend: FastAPI

The architecture is offline-first — small LLMs run on mid-range smartphones without requiring internet connectivity.

Challenges we foresee

Building a robust analogy knowledge base that resonates across diverse rural contexts
Balancing simplification with accuracy (can't oversimplify financial risks)
Handling regional variations in language and cultural references
Optimizing LLM inference for low-resource devices

What we learned so far

Financial literacy isn't about explaining finance better — it's about connecting new concepts to existing knowledge. Rural Indians already understand risk, investment, and returns through farming. We just needed to bridge that gap.