Back to Portfolio
Case StudyPrivate Beta

Sprich

A B2B language-learning platform built around one promise: your team speaks in week one, not month six — with CEFR progress tracked automatically.

11 languages|54 unit tests|Per-seat B2B billing

The Problem

Consumer language apps optimize for streaks, not for getting a working professional conversational quickly. L&D teams buying language training need something different: real speaking practice from day one, objective progress they can report on, and the billing, admin, and compliance controls a company actually requires before it rolls a tool out to staff.

The Approach

Lead with conversation. Sprich puts learners straight into AI-driven speaking practice — voice in, voice out, with corrections captured on every turn — and tracks placement against the CEFRframework so progress is measurable rather than gamified. A free tier stays open for individuals; the business model is per-seat for teams.

Around the product sits a full B2B layer: organizations, members, email invites, a team admin dashboard, and Stripe per-seat checkout that routes subscriptions to the org or the individual based on metadata.

The Build

Everything below is built and on main:

AI conversation practice (11 languages)
Voice in (Deepgram) + streaming voice out (ElevenLabs)
CEFR placement + ongoing reassessment
Corrections stored per message
Public /try demo — no signup, rate-limited
Org / member / invite data model
Per-seat Stripe billing with adjustable quantity
Team admin dashboard with KPI cards
Weekly digest — KPI engine + AI narrative
PII redaction before any model call

Reliability was treated as a feature: a five-layer recovery pipeline wraps every upstream call (chat, demo, transcribe, TTS) with per-provider circuit breakers, so a model or voice outage degrades to a clean message instead of a crash. Compliance v0 redacts emails, phones, SSNs, Luhn-valid cards, and IPs before anything reaches the model, with per-org retention enforced by a daily cron.

The Result

Sprich is feature-complete for a sales-ready v1 — 54 passing unit tests, a green Turbopack build, marketing surface reframed for L&D buyers, and a white-label demo for prospects. It's the pattern source for the error-recovery, metering, and reporting-dashboard work I reuse across other products.

Stack

Next.js 16React 19TypeScriptPrismaSupabaseStripeGeminiDeepgramElevenLabsnext-intlTailwind CSS

Want something like this?

I ship production-ready applications fast. Let's talk about your project.

FL

FormationLabs

AI Assistant

Hey! I'm the Formation Labs AI assistant. How can I help you today?

Quick questions:

Powered by FormationLabs