FinKar

• 

  Agent Architecture

• 

  Live Stock Market Analysis Page

• 

  Stock Market Learning module

• 

  Home page

• 

  Stock Market Learning Page

• 

  Financial Tracker Page

Inspiration

• Large section of Indians are financially underserved, not due to lack of interest, but poor tool design
  
• Most platforms assume fixed monthly salaries and English-only usage
  
• 500M+ Indians prefer regional languages and many earn via irregular/gig incomes
  
• FinKar was inspired to align financial guidance with real Indian financial behavior

What it does

• AI-powered personal financial coach (not just an expense tracker)
  
• Supports 14 Indian languages with automatic language detection
  
• Adapts to irregular income patterns (freelancers, gig workers)
  
• Delivers contextual and proactive financial advice
  
• Uses secure, consent-based real financial data to help users:
  • Understand spending patterns
    
  • Plan savings
    
  • Make informed financial decisions
    

How we built it

• Multi-agent AI architecture orchestrated using LangGraph
  
• Four specialized agents:
  • Query planning
    
  • Context retrieval
    
  • Financial reasoning
    
  • Proactive coaching
    
• RAG pipeline with Pinecone for grounded financial responses
  
• Groq LLMs for low-latency inference
  
• RBI-regulated Account Aggregator integration (via Setu)
  
• Backend: FastAPI with asynchronous pipelines
  
• Frontend: Cross-platform mobile app using React.js + Ionic

Technical Innovation

• Multi-agent AI financial system built using LangGraph, enabling modular reasoning instead of monolithic LLM responses
  
• Agent specialization for planning, retrieval, financial reasoning, and proactive coaching improves accuracy and reduces hallucinations
  
• Retrieval-Augmented Generation (RAG) powered by Pinecone, grounding responses in verified financial knowledge
  
• Low-latency inference using Groq LLMs, ensuring real-time user interactions
  
• Consent-driven real financial data access via RBI-regulated Account Aggregator framework (Setu)
  
• Irregular income-aware financial reasoning, specifically designed for gig workers and non-salaried users
  
• Automatic multilingual intent detection and response generation across 14 Indian languages
  
• Proactive coaching engine that surfaces insights without explicit user queries
  

Challenges we ran into

• Coordinating multiple AI agents while maintaining low latency
  
• Preserving intent across multilingual inputs
  
• Handling consent flows and data normalization for Account Aggregators
  
• Balancing depth of financial reasoning with responsiveness

Accomplishments we're proud of

• Built a complete end-to-end mobile application
  
• Integrated real financial data securely
  
• Deployed a working multi-agent AI system
  
• Enabled financial coaching in 14 Indian languages

What we learned

• Model strength alone is insufficient — architecture and retrieval are critical
  
• Multi-agent systems require careful orchestration
  
• Fintech AI must prioritize security, consent, and compliance

What's next for FinKar

• Investment and credit guidance
  
• Long-term financial and goal planning
  
• Improved personalization through continuous learning
  
• Voice-based interactions for accessibility
  
• Scaling to more financial institutions and user segments

Built With

• account-aggregator-(setu)
• fastapi
• groq-llms
• ionic-framework
• javascript
• langchain
• langgraph
• multi-agent-systems
• multilingual
• natural-language-processing
• pinecone
• python
• react.js
• retrieval-augmented-generation-(rag)
• sql
• sqlite