🧠 Hybrid Multilingual RAG · Ensemble Sentiment · Economic Forecast

No index built yet.

🎤 Speak your question — get a spoken answer

🧠 Hybrid Multilingual RAG Framework

🤖 Models Used

• FinBERT (ProsusAI) — Financial sentiment — 40%
• XLM-RoBERTa (CardiffNLP) — Multilingual sentiment — 30%
• Economic Lexicon — Domain-specific keywords — 30%
• MiniLM-L12 — Multilingual embeddings + FAISS
• ms-marco-MiniLM — Cross-encoder reranking
• Whisper-small — ASR
• Llama-3.3-70B via Groq — Response generation

📈 Forecasting

• Baseline: ARIMA(1,1,1)
• Enhanced: SARIMAX + Ensemble Sentiment (n_test = 3)
• Quarterly: U-MIDAS + Sentiment Lags
• Tests: ADF · Granger Causality · Diebold-Mariano
• Data: World Bank API + IMF Quarterly CSV

📈 Economic Forecast

Option A: Yearly SARIMAX (World Bank API) | Option B: Quarterly U-MIDAS (IMF CSV)

🅰️ Option A — Yearly Forecast (SARIMAX + Ensemble Sentiment)

Uses World Bank API. Evaluates on last 3 years.

🅱️ Option B — Quarterly Forecast (U-MIDAS + Sentiment Lags)

1. Build PDF index in Tab 1 first
2. Upload raw IMF CSV
3. Press button below