Vernacular FD Advisor
EXECUTIVE SUMMARY
Vernacular FD Advisor is a multilingual, RAG-powered financial assistant that helps Indian users understand, compare, and plan Fixed Deposits using trusted, source-backed knowledge. Instead of relying on generic LLM responses, the system retrieves relevant financial information from authoritative sources such as RBI, SEBI, DICGC, and bank documentation before generating grounded, vernacular-first guidance.
PROBLEM STATEMENT
Fixed Deposit information in India is fragmented across multiple websites, filled with banking and regulatory jargon, difficult to compare, and largely inaccessible to vernacular-first and first-time savers. Generic AI assistants are unreliable in finance because they hallucinate and provide non-verifiable advice. Vernacular FD Advisor addresses this by delivering trusted, understandable, multilingual financial guidance grounded in retrieved knowledge rather than open-ended generation.
INVESTIGATIVE FINDINGS
- ▸Financial information is scattered across portals and buried under complex regulatory jargon
- ▸Vernacular and mixed-language users are underserved by English-only retrieval systems
- ▸Generic LLM responses risk hallucination, making them unsuitable for financial advisory
- ▸Source-backed retrieval (RBI / SEBI / DICGC) is required to establish trust in financial guidance
BEHAVIORAL ANALYSIS
The system is designed as a trust-first financial assistant rather than an open-ended chatbot. By grounding every response in retrieved, verifiable sources and guiding users through a structured FD decision journey, it prioritizes correctness, transparency, and accessibility over conversational flexibility. The vernacular-first design ensures financial literacy reaches users beyond the English-speaking demographic.
TECHNICAL EXECUTION
User queries pass through language understanding and are embedded using multilingual-e5-large embeddings. Semantic retrieval against a Supabase pgvector store surfaces relevant financial knowledge, which is injected into the LLM (Groq, openai/gpt-oss-120b) to generate structured, source-backed multilingual answers. Voice queries are transcribed via Groq Whisper (whisper-large-v3-turbo). An automated n8n ingestion pipeline fetches trusted sources, cleans content, applies SHA256 hash-based change detection, performs semantic chunking, and stores multilingual embeddings to keep the knowledge base fresh and duplicate-free.
SYSTEM ARCHITECTURE
OPERATIONAL FEATURES
- ✓Multilingual AI assistant (English, Hindi, Hinglish, Marathi, Gujarati, Tamil, Bhojpuri)
- ✓Guided FD decision flow covering bank, amount, tenure, and FD type selection
- ✓Interactive FD Explorer with filtering by amount, tenure, institution type, and risk
- ✓Bank comparison workspace for side-by-side rate and safety analysis
- ✓Built-in FD return calculator with compounding support
- ✓Voice input via Groq Whisper transcription
- ✓Automated n8n knowledge ingestion with SHA256 change detection
TECHNOLOGICAL ASSETS
CURRENT STATUS
Working hackathon prototype (Hack to the Future) with end-to-end multilingual FD advisory flow, implemented RAG pipeline, and automated ingestion architecture. FD opening is currently a guided simulation; live bank API integrations and real-time rate syncing are planned.
CONCLUSION
Vernacular FD Advisor demonstrates how RAG and multilingual retrieval can deliver trusted, source-backed financial guidance to underserved vernacular users. By grounding responses in authoritative sources and guiding users through a structured FD journey, the system offers a production-inspired blueprint for accessible, hallucination-resistant financial advisory.