RiskNarrate

• 

  Risk score summary & trend

• 

  Anomaly detection

• 

  Gemini‑powered risk explanations

• 

  6‑month risk forecast & CIBIL‑like credit estimate

• 

  What‑If Simulator that predicts how actions affect future risk

• 

  Fairness & Explainability

RiskNarrate

✨ Inspiration Financial risk analysis is traditionally slow, manual, and difficult for non‑experts to interpret. We wanted to create a system where risk explains itself combining structured financial signals with the reasoning power of Gemini 3. This led to RiskNarrate, an AI copilot that can read customer histories, understand trends, and narrate them in simple language.

🛠️ How We Built It RiskNarrate is built using: Streamlit for the interactive dashboard Pandas + Matplotlib for data processing and visualization Custom risk engine using formulas such as: risk_score=w1⋅u+w2⋅m+w3⋅a where u is utilization, m is missed payments, and a is anomaly flags Gemini 3 API (google‑genai) for: Risk explanations Lender‑fit reasoning ReportLab for a fully generated PDF report A What‑If simulator that applies adjustments: new_utilization=current×(1−r) new_missed=max(0,m−x) We structured the app so the deterministic models compute numeric risk, and Gemini 3 narrates the insights.

📚 What We Learned Building RiskNarrate taught us: How to design compact context windows for Gemini for fast reasoning How to blend rule‑based scoring with generative intelligence How to optimize API latency by reducing token size How to generate enterprise‑style PDFs programmatically How to deploy a generative AI app publicly on Streamlit Cloud Gemini 3’s ability to turn structured CSV data into natural‑language insight was a major unlock.

🚧 Challenges We Faced We encountered several real‑world issues: Slow analysis speed until we optimized context and caching Streamlit reruns causing latency spikes during What‑If simulations Deployment errors due to missing SDKs (google‑genai) on Streamlit Cloud Chart + PDF layout tuning to keep the final report readable Designing a clean UI while keeping multiple components (forecast, credit score, anomalies, what‑if) in sync Every challenge helped us refine and stabilize the final application.

🎯 Conclusion RiskNarrate shows how combining deterministic models with Gemini 3’s reasoning unlocks a new kind of explainable, transparent, and actionable financial risk intelligence all in a simple, interactive experience.

Built With

• github
• google-gemini
• googlegeminiapi
• matplotlib
• numpy
• pandas
• pydantic
• python
• reportlab
• scikit-learn
• streamlit
• streamlitcloud
• vscode