AI-Powered Chatbot for Smarter Insurance Sales — Fidelidade (Case 4)

NOVA IMS — MSc in Data Science and Advanced Analytics
Course: Business Cases with Data Science (2025)
Client: Fidelidade · Instructor: Prof. Nuno António

---

🎯 Objective

Develop an AI assistant to help Fidelidade insurance agents provide faster, more accurate, and more educational support to clients.

The chatbot delivers:

• Real-time information about My Savings and PPR Evoluir
• Financial literacy explanations (based on Fidelidade Academia FAQs)
• Comparisons with competitor financial products
• Links to regulatory and tax resources (ASF, CMVM, BdP, etc.)

---

⚙️ Approach

• Data ingestion: PDFs, FAQs, and official product web pages
• Preprocessing: text chunking, embedding generation, vector indexing
• Model: OpenAI API (GPT-4 / Assistants API) or Gemini Vertex Agent Builder
• Retrieval: similarity search over embeddings + context re-ranking
• Interface: conversational prompt pipeline with context memory

---

📚 Training Sources

1. Product Information: My Savings, PPR Evoluir
2. Competitor Products: CGD, Santander, Millennium, BPI, Optimize, etc.
3. Financial Literacy: FAQS_FIDELIDADE_ACADEMIA.pdf
4. Tax & Regulation: ASF, BdP, CMVM, TodosContam
   (see data/training_docs/ for organized content)

---

🧠 Evaluation (per Fidelidade’s brief)

Criterion	Description
Accuracy & Clarity	Answers product questions precisely and politely
Coverage of FAQs	Handles literacy and insurance topics
Context Awareness	Understands follow-up questions
Use of Resources	Cites official PDFs or URLs
Sales Support	Highlights product advantages

---

📂 Repository Structure