Optimate

optimate-demo-1758064916274.mp4

Optimate is an AI-powered underwriting dashboard that transforms raw insurance policy data into actionable insights, helping underwriters make faster, smarter, and more confident decisions.

---

Elevator Pitch

“Optimate: AI dashboards, insights, and reinforcement learning for smarter underwriting.”

Optimate reimagines the underwriting process by blending AI, reinforcement learning, real-time tracking, and intuitive visualizations into one seamless platform.

---

Features

• Top Matches → View the top 10 submissions that best align with appetite guidelines.
• Explainability → Get detailed AI-driven justifications for why a policy is prioritized.
• Approve/Decline Actions → Take decisions directly from the dashboard.
• Reinforcement Learning → Provide satisfaction feedback so the AI improves over time.
• Visual Analytics → Explore plots, live tracking, and interactive heatmaps of U.S. states colored by appetite and risk scores.
• Contextual AI Chatbot → Two-way conversations with the AI at the individual policy level, enabling clarifications and co-decision making.

---

How We Built It

• Frontend & Deployment:

  • Next.js (App Router) + TailwindCSS + shadcn/ui
  • Deployed on Vercel for fast CI/CD and scalability.
  • Auth0 for secure authentication and role-based access.

• Data Layer:

  • Policy data ingested from the Federato API.
  • Enriched in Python with risk scores, appetite scores, and account-level analytics.

• AI & Insights:

  • CohereAI LLMs + Retrieval-Augmented Generation (RAG) to explain why submissions are in-appetite.
  • Reinforcement learning loop powered by underwriter feedback (approve/decline, satisfaction scores).
  • Context-aware AI chatbot at the policy level.

• Storage & State Management:

  • AWS DynamoDB for user preferences, chat history, cached embeddings, and saved searches.

• Visualization:

  • Built with Recharts and D3.js to show:
    • Interactive plots
    • Risk & appetite heatmaps
    • Portfolio leaderboards

---

Inspiration

We were inspired by Federato’s challenge:

“Reimagine the RiskOps landing page experience… intelligently surface a curated set of submissions that are most aligned with the carrier's appetite—enabling faster and more effective underwriting decisions.”

Underwriters often juggle complex and conflicting factors like TIV, loss ratios, building age, and state regulations. Optimate was built to simplify this decision-making by combining AI, reinforcement learning, and visual storytelling into one dashboard.

---

Challenges We Ran Into

• Understanding new insurance/underwriting concepts like TIV, appetite guidelines, and loss ratios.
• Handling merge conflicts in a 4-dev hackathon team under time pressure.
• Designing a reinforcement learning loop that balanced rule-based scoring with AI reasoning.
• Securing AWS DynamoDB with the right IAM roles and schema design.
• Optimizing visualizations for responsiveness while handling large datasets.

---

Accomplishments

• Delivered a production-ready product within the hackathon timeframe.
• Rapidly learned and integrated CohereAI, Auth0, DynamoDB, and shadcn/ui.
• Implemented reinforcement learning foundations for underwriting feedback loops.
• Built heatmaps, live tracking, and plots that made underwriting metrics intuitive.

---

What We Learned

• Clear team communication is crucial under time pressure.
• How to design and deploy a RAG pipeline with Cohere.
• Practical application of reinforcement learning in a real-world use case.
• Insights into the insurance industry and underwriters’ real challenges.
• How to take a vague problem statement and ship a polished end-to-end solution.

---

What’s Next

• Deeper AI → Fine-tune LLMs for nuanced underwriting strategies.
• Dynamic Data Feeds → Add weather, property valuation, and catastrophe models.
• Explainability → Richer visual cues (confidence intervals, score breakdowns).
• Collaboration → Shared views, notes, and histories for underwriting teams.
• Mobile → Extend Optimate to iOS/Android for underwriting on the go.
• Enterprise-Ready Infra → Multi-region cloud deployment, audit trails, and monitoring.

---

🛠️ Built With

Next.js, Vercel, TailwindCSS, shadcn/ui, Auth0, CohereAI, Python, AWS DynamoDB, Federato API, Recharts, D3.js, Retrieval-Augmented Generation (RAG), Reinforcement Learning, OAuth 2.0

---

🔗 Links

• Deployed App → Optimate on Vercel
• Hackathon Challenge → Federato RiskOps Hackathon

---

Team

Built with ☕ and teamwork by four developers at Hack the North 2025.