Mikhail IvanovAll case studies / AI Assistants & Agents
MVP platform where AI creates teaching materials, with a quality judge and data isolation
Industry: Education startup (EdTech)Service: AI assistants and agents
550+ testsindependent verifier run, types passed without errors
0 leaksbetween organizations, verified on a real Supabase instance
8 sectionsin each material, two-level safety check
Before
- The platform generates activity cards for preschool and elementary school educators
- The client specified the stack: Supabase, Next.js, Stripe, OpenAI, Vercel
- Plans for individual educators and programs with teams; directors can see their educators' history
What we built
- Material generation by age, topic, and conditions with a two-level safety check
- Quality measurement: a code gate catches missing content, an LLM judge scores against a rubric using seeded violations
- Organization data isolation with RLS policies in PostgreSQL
- Stripe subscriptions, emails via Resend, PDFs in color and black and white
- Deployment on Vercel, observability through Sentry and PostHog
How it works
Educatorage, topic, conditions
→
Next.js APIquota and profile
→
GenerationOpenAI, 8 sections
→
Checkssafety and quality judge
→
Resultcard and PDF
Result
- First stage accepted: link works, emails are delivered, login and password reset verified
- Core MVP work paid for
- The client reserved three more stages
- Generation and PDF work in the acceptance version of the second stage, confirmed by the client
Technology
What matters in a project like this. We first tested isolation with a local stub, but only a run on a real Supabase instance was accepted as proof. A separate verifier caught a false-green contractor report, so readiness is now confirmed by a second reviewer.
Have a similar project?
Describe it in two lines in Telegram. We'll explain how we'd approach it, how long it would take, and what it would cost.
Similar case studies
Auto repair shop
A bot sends the shop owner a daily summary, while AI finds discrepancies in the numbers
every daydaily summary for the owner and mechanic workload in the morning
Auto parts marketplace
LLM parts search audit cuts query cost by 16.7% without losing accuracy
−16.7%cost per query: $0.1353 → $0.1128
SaaS for local businesses: review responses
SaaS collects reviews from 8 platforms and drafts AI responses for each business niche
8 platformsreviews collected from them