Waypoint
The Roadtrip Method, Turned Into a Tool · AI & Language Learning
Overview & The Problem
Waypoint is a language-learning web app that operationalizes the immersion-based "Roadtrip Method" into structured Input / Output / Maintenance sessions, with an AI-assisted speaking-and-writing loop that turns your own mistakes into spaced-repetition flashcards.
Self-directed language learners are told to immerse, speak, and review — but no single tool sequences those activities into a coherent daily practice. Flashcard apps drill isolated cards with no connection to real output; there is no bridge from "I tried to say something and got it wrong" to "that exact gap is now a card I review." Learners are left to stitch together Anki, a voice recorder, a notebook, and a tutor by hand, and most abandon the loop before it compounds.
My Role
Solo designer and full-stack developer. I adapted the Roadtrip Method (an existing immersion pedagogy) into a concrete session model, and built the entire app end to end:
- The React/TypeScript client and its session UI.
- The Supabase data model and authentication.
- The two AI edge functions — transcription and correction/flashcard mining.
- The SM-2 spaced-repetition engine.
- The story-speaking practice loop that ties it all together.
Solution & Approach: The Story-Speaking Cycle
Waypoint structures practice around the Roadtrip Method's three pillars — Input, Output, and Maintenance — inside timed session blocks. Its centerpiece is the Story-Speaking cycle: the learner records themselves telling a story from a prompt, Whisper transcribes it, and a three-column review workbench (what you said / what you meant / the right way) has Claude generate corrections and mine the gaps into flashcards.
The learner then briefly drills those cards and re-records the same story — repeating the cycle until the vocabulary and narrative stick. Cards flow into an SM-2 review queue, so today's spoken mistake becomes tomorrow's scheduled review.
A parallel Writing flow uses the same review-and-mine workbench, and dedicated Speaking and Writing pages let learners search past sessions, rename them, study a single session's mined cards, or start an exercise standalone.
Technologies Used
- Language: TypeScript
- Client: React 19 + Vite, Tailwind CSS v4
- State: React Router, TanStack Query (server state), Zustand (session state)
- Backend: Supabase (Postgres, Auth, Storage, Edge Functions)
- Speech-to-text: OpenAI Whisper, via a Deno edge function
- Correction + flashcard mining: Anthropic Claude, via a Deno edge function with structured JSON output
- Scheduling: SM-2 spaced-repetition
- Delivery: PWA (installable, offline-aware)
Impact & Intent
Waypoint is in active development. Its aim is to close the loop between production and review that self-study normally leaves open — making the learner's own output the source of their spaced-repetition material, inside a session structure that reflects an established immersion method rather than ad-hoc drilling. The outcomes below are design intent rather than measured results:
Language learners:
- One tool that sequences immersion, speaking/writing output, and review — instead of manually juggling a flashcard app, a recorder, and a notebook.
- Flashcards sourced from the learner's own mistakes, so review time targets exactly the gaps that surfaced during real production.
The method:
- Turns the Roadtrip Method's Input / Output / Maintenance pillars into a concrete, timed session flow a learner can actually follow day to day.
- The Story-Speaking cycle makes "repeat the same story until it's fluent" a first-class, guided loop rather than a vague instruction.
The app / engineering:
- A clean layered architecture (domain entities → repository interfaces → Supabase implementations → React Query hooks) that keeps UI code free of direct database access.
- AI keys never touch the client: transcription and correction run in Supabase edge functions, which also expose availability probes so the UI degrades gracefully when AI is unconfigured.
The pedagogical framework is adapted from the existing "Roadtrip Method" immersion approach; Waypoint is the software implementation of it, not the method's originator.
Visuals & Links
Source for the Waypoint client and edge functions
The Story-Speaking cycle: record → transcribe → correct → mine flashcards → study → retake.