Real-Time Plan Matching with Streamlit, OpenAI Embeddings, and Vector Search

Executive Summary

Statistique partnered with CardinalSB to build an AI-powered insurance recommendation system that gives agents real-time access to the best insurance plans across multiple carriers. In just three months, the tool went from concept to deployment—built using Streamlit, FastAPI, and OpenAI embedding models for document search.

Now live, the solution is used by 100+ agents, supports 7 integrated providers, and generates personalized insurance quotes in seconds instead of minutes. The result: faster quoting, improved accuracy, and a major boost in agent productivity.

The Challenge

Disparate Provider Criteria
Carriers shared underwriting and pricing guidelines as PDFs or Word documents—with inconsistent formats, terminology, and structure.

Manual Comparison Overhead
Agents spent over 10 minutes per client manually parsing documents or checking with supervisors to find eligible plans.

Inconsistent Recommendations
Without a unified engine for eligibility logic, agents often delivered inconsistent or suboptimal plan matches—leading to lost opportunities and a degraded client experience.

The Solution: AI-Powered Real-Time Insurance Matching

🧾 Lightning-Fast Intake with Streamlit

A sleek, responsive form – built in Streamlit, lets agents capture client details (age, health conditions, location, etc.) in under 30 seconds.

• Smart validation ensures accuracy from the start

• Designed for speed, clarity, and minimal training time

• Optimized for busy agents juggling high volumes

Seamless Insurance Document Automation

Every week, the system ingests and updates carrier documents – PDFs, Word files, or scanned guides.

• OCR + PDF + semantic chunking breaks complex docs into structured data

• Transformed into OpenAI-powered vector embeddings for meaning-based retrieval

• Indexed in Pinecone for lightning-fast vector search across thousands of eligibility criteria

Intelligent Query Generation

A custom-built engine converts intake responses into no noise, time-aware natural language queries.

• Maintains precision on diagnosis dates, durations, and medical details

• Enables high-accuracy matches without clunky rule trees

• Designed to think like an underwriter, minus the delays

Automated Documentation & Agent Delivery

Every time a recommendation is generated:

• A PDF summary of the inputs and results is auto-created

• Stored directly to a secure Google Drive folder

• Emailed instantly to the agent for easy documentation, recordkeeping, or client handoff
  → No more screenshots, copying, or manual uploads

⚙️ Effortless CRM Integration

The tool is embedded directly inside CardinalSB’s CRM using an iframe.

• No context switching – recommendations appear where agents already work

• Zero training required

• Maximum adoption and daily usage from day one

Implementation Highlights

Timeline:

• 3 months total (build + test)

• Live by end of Month 3

Scale & Scope:

• 7 insurance providers integrated

• 100+ agents onboarded during rollout

• Designed to onboard new carriers in <1 week

Tech Stack:

• Frontend: Streamlit + React widgets

• Backend: FastAPI, hosted on AWS ECS with autoscaling

• Vector Search: Pinecone (Weaviate compatible)

• Embeddings: OpenAI text-embedding-3-small

• Security: HIPAA-compliant architecture with full encryption and isolated VPC

Results & Impact

⏱️ 10x faster quote generation
What took over 10 minutes now happens in under 10 seconds.

🎯 Improved plan-match accuracy
No more guesswork or supervisor dependence—just precise, AI-powered results.

🚀 3–4× agent productivity
Agents can now serve more clients, faster, with better confidence.

📈 Scalable infrastructure
New providers can be added in days, supporting CardinalSB’s expansion roadmap.

💬 Early feedback = high satisfaction
Agents report smoother workflows and better client conversations.

“Honestly, this AI agent has been a huge win for us. Our agents get accurate quotes in seconds, without having to dig through documents or chase down supervisors. It’s made their jobs easier and our client experience smoother. We couldn’t be happier.”
– Nick Dale, CEO, CardinalSB

 

Key Takeaways

✅ End-to-end AI automation: From form capture to vector-based retrieval, built for speed and scale
✅ Document parsing at scale: Automates messy underwriting PDFs into searchable knowledge
✅ Time-aware queries: Intelligent query engineering ensures retrieval precision
✅ Seamless UX = adoption: CRM-embedded design drives usage and ROI
✅ Future-ready design: Modular build supports predictive features and consumer-facing rollouts

Next Steps

• Expand provider network to 15+ carriers by Q4

• Launch churn modeling and premium optimization modules

• Deploy client-facing portal for self-service insurance matching

• Integrate computer vision to allow agents to simply photograph medications—the model will recognize them, infer their uses, and instantly determine plan eligibility based on provider criteria

Ready to Transform Your Operations?

This project is a testament to what’s possible when deep expertise meets cutting-edge AI. Statistique didn’t just build a tool – we delivered a complete transformation of CardinalSB’s quoting workflow, from document chaos to intelligent automation.
If you’re ready to modernize your operations with AI, automation, or custom data solutions, get in touch with us here. We’d love to help.