The AI Platform powering NatCat Property Insurance
Overview
ResiQuant is a San Francisco-based, seed-stage startup founded by Stanford PhDs and backed by Pear VC, LDV Capital, and Foothill Ventures. The platform empowers property insurers to underwrite high-risk buildings using AI-powered document parsing, structural data extraction, and automated risk scoring.
Faced with tight investor timelines and a fragmented prototype, ResiQuant partnered with Spartan to design and deliver a production-grade MVP + fast.
The Challenge
Traditional insurance underwriting relies on manual review of ACORD PDFs and unstructured property submission documents: a slow, error-prone, and non-scalable process.
Key challenges included:
- Unstructured Inputs: Inconsistent PDF formats, especially with tabular building data.
- High Accuracy Requirement: 90%+ extraction precision needed to drive underwriting logic.
- Time Pressure: 2-week deadline for a live demo to secure investor interest.
- Tech Debt & Infra Bottlenecks: Initial system suffered from memory leaks, poor job orchestration, and inflated infra costs.
Spartan’s Solution
Spartan deployed a compact, full-stack team (backend, frontend, AI, DevOps) to own delivery of a functional MVP within 4 weeks.
Key Deliverables
- End-to-End AI Workflow
Document ingestion → parsing → validation → risk assessment.
- PDF Parsing System
Benchmarked Azure Document Intelligence vs. Camelot, Textract, and PdfPlumber, and selected Azure Document Intelligence for the highest table accuracy. Augmented with preprocessing (resolution upscaling, section splitting), fallback parsers, and few-shot prompting for LLM-based validation.
- Backend & Job Orchestration
Async processing pipeline (SQS + microservices), confidence scoring, retry logic, and secure APIs for structured submission flow.
- Frontend & User Flows
Property lookup (Google Maps + Mapbox), PDF uploader, image gallery, and risk profile visualizations.
- Infra Foundation
AWS EKS, S3, PostgreSQL, Redis, OpenSearch, Jenkins, Terraform, Azure OpenAI, Slack APM.
Results
| Metric | Impact |
|---|---|
| Demo Delivered | < 2 weeks → enabled VC outreach & conference demos |
| Table Extraction Accuracy | >90% (Azure Doc Intelligence + prompt tuning) |
| Processing Speed | Improved by ~50% |
| Infra Cost | Reduced by 60–70% |
| System Stability | Production-ready orchestration in 1.5 months |
Why Spartan
- Full-Stack Velocity: Shipped complex AI + infra system in weeks, not months.
- Tech Rigor under Time Pressure: Delivered investor-grade demo without sacrificing accuracy or scale.
- Deep Collaboration: Embedded with founders and PMs (async loops, daily Slack check-ins, and hands-on design reviews).
- AI-Native Thinking: Combined LLM, OCR, and cloud-native infrastructure to unlock real business value.
