Multi-Tenant Vector Isolation for a B2B Knowledge Assistant
Overview
What this challenge is about.
Build a small proof-of-concept in your chosen vector store (Pinecone or Qdrant — pick one and justify) that supports 10 simulated tenants with 1,000 vectors each. Implement the recommended isolation pattern. Then build a leakage test suite: for each tenant, run a query that intentionally matches a high-similarity vector belonging to another tenant, and assert zero cross-tenant results across 200 such probes. Measure per-tenant query latency (p50/p95) and ingestion throughput. Write a 4-page design document covering the pattern decision, the threat model, blast-radius if isolation fails, and how the pattern scales to 300 tenants. Include a clear answer to: when does pattern (1) flip to (3)?
The Brief
What you'll do, and what you'll demonstrate.
Pick and prove a tenant-isolation pattern for a multi-tenant vector store that survives 200 adversarial cross-tenant probes and scales credibly to 300 tenants.
Earning criteria — what you'll demonstrate
- Compare multi-tenant patterns (per-collection, filtered, per-database) on isolation, cost, and operability
- Build a threat model for a vector store handling customer-confidential data
- Write automated leakage tests that survive code refactors
- Project tenant-scale economics for an enterprise SaaS product
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
AI Solutions Architect
Designing tenant isolation for a vector workload, defending it to procurement, and projecting it to 300 tenants is the AI solutions architect's signature deliverable for B2B AI startups.
This challenge sharpens
- multi-tenant-isolation
- vector-databases
- threat-modeling
MLOps Engineer
Owning the leakage test suite in CI and the per-tenant latency story is the MLOps work that keeps a multi-tenant retrieval system honest as it scales.
This challenge sharpens
- multi-tenant-isolation
- vector-databases
- evaluation
AI Safety Researcher
Building an adversarial probe suite against a retrieval system to find cross-tenant leakage is directly transferable to red-team work on production AI systems.
This challenge sharpens
- threat-modeling
- evaluation
- security-engineering