All case studies
B2B SaaS · Legal-tech
Multi-tenant LLM platform for an industry SaaS.
Powered 3 customer-facing AI features across 80 tenants on a single stack.
LLM platformMulti-tenantStreaming
Outcomes
By the numbers.
80Active tenants
3AI features
−42%Cost per req
Challenge
The problem we were brought in to solve.
The product team wanted three different AI features but couldn't justify three separate stacks, three sets of evals, or three sets of cost controls per tenant.
Approach
How we built it.
- Built a per-tenant model gateway with rate limits, cost ceilings, and audit logs.
- Standardized on a streaming SSE protocol shared by all features.
- Centralized prompt management with A/B testing and per-tenant overrides.
- Added an eval pipeline that runs nightly against a fixed regression set.
Stack
The technology we used.
OpenAI
Anthropic
Vercel AI SDK
Postgres
Redis
LangSmith
Related services
Have a similar problem to solve?
Tell us about your project. We respond within one business day.