Suppliers that look independent can share one hidden upstream risk.
A full supply chain platform with cost workflows and explainable supplier risk scoring.
80+ endpoints, and a critical (P0) bug caught before launch.
The problem
Supplier risk platforms typically evaluate the supplier directly in view, then stop. When multiple ostensibly independent suppliers share a single upstream dependency, conventional tools report several separate, healthy scores — until that shared source fails, and every dependent supplier fails simultaneously, without warning.
Who it is for
- Procurement and supply chain organizations requiring visibility beyond immediate, direct supplier relationships
- Operations leadership accountable for business continuity risk, currently reliant on manual, spreadsheet-based supplier tracking
- Mid-market manufacturing and logistics organizations requiring risk intelligence without enterprise-scale platform investment
Market context
This is an established, well-capitalized category with serious institutional competitors — Resilinc (450,000+ pre-mapped suppliers), Sphera (a dedicated N-Tier visibility product), Everstream Analytics, and Interos (400M+ entity graph), alongside legacy enterprise incumbents SAP Ariba, Oracle SCM, and Blue Yonder.
The gap it closes
Competing platforms differentiate primarily through database scale — hundreds of thousands of pre-mapped supplier relationships. SCIP is architected instead for comprehension and actionability of multi-tier dependency risk within an organization's own supplier network — rendering the dependency chain visually explicit, with sole-source exposure flagged as a discrete signal rather than obscured within a composite score.
How it works
- Supplier data is imported via CSV — operational without requiring existing ERP integration
- The system evaluates direct supplier risk, then traces upstream dependencies of that supplier
- Sole-source dependencies — critical inputs without backup sourcing — are flagged as explicit, discrete warnings
- A visual dependency chain diagram renders the precise location of concentrated risk
- Scenario modeling enables evaluation of hypothetical risk changes without affecting production data
What makes it different
- Multi-tier visibility without dependency on a pre-built database — operational from day one against your own supplier network
- Sole-source exposure isolated as an explicit signal, never obscured within an aggregated composite score
- Empirically validated, not asserted — stress-tested against irregular real-world data, with encoding defects identified and resolved during that validation process
Architecture
System Overview
Component Breakdown
| Spring Boot API | Enterprise-grade transactional core — orders, suppliers, inventory, auth |
| FastAPI AI Service | Isolated Python service for ML workloads |
| Cascading Risk Engine | Traces the supplier dependency graph, flags sole-source exposure explicitly |
| HNSW Vector Index | Fast similarity search over supplier/document embeddings |
Data Flow
- Supplier data enters via CSV import — no ERP integration required to start
- IsolationForest scores each supplier's direct risk
- The cascading engine traces upstream dependencies, computing effective risk beyond the direct score
- Sole-source dependencies flagged as an explicit, separate signal
- Claude AI generates a plain-English risk explanation
- The dependency chain renders visually on the Supplier Detail page
Technology Stack
| Backend core | Spring Boot 3 (Java 17) | Enterprise-grade REST robustness |
| AI/ML service | Python FastAPI | Real ML ecosystem the JVM doesn't match |
| Vector search | pgvector + HNSW | 28ms → 1.25ms, confirmed against 8,000+ documents |
| AI explanations | Claude AI | Natural-language grounding for risk scores |
Real Design Trade-off, Documented
The Java/Python split was deliberate — Java kept for transactional API maturity, Python kept because its ML ecosystem is stronger. The accepted trade-off: two runtimes to deploy and maintain, rather than one.
Performance, Measured Not Estimated
- Vector search: 28ms → 1.25ms, a real ~22x improvement, confirmed against a live database
- 80+ real REST endpoints, tested via 41 Playwright test cases across 8 modules
- A real P0 bug found pre-launch: BCrypt.matches(null, hash) returning true
See SCIP working
A guided walkthrough, one-on-one. We usually reply within one business day.