ConceptLens — Developer Learning Platform

Learners could generate a batch of AI‑written explanations, but there was no mechanism to verify whether they could actually articulate the material. The process stopped after the initial content creation, leaving gaps in understanding unaddressed. Without ongoing assessment, learners never knew which concepts they truly mastered.

I built ConceptLens on a Next.js 15 front‑end with Supabase handling authentication and data persistence. LangChain.js orchestrates prompts to the Groq gpt‑oss‑120b model, keeping latency low while controlling costs. Inngest powers background workflows that schedule follow‑up interviews and rubric checks after each study session. The core loop lets a learner edit AI‑drafted tracks, dive deeper via chat, then face an AI interviewer that grades responses against a rubric generated alongside each question.

When I first examined existing AI‑driven learning tools, I saw a clear gap: they could spin out explanations at speed, but they never closed the feedback loop. Students walked away with a stack of content they had never spoken aloud, and there was no way to surface the concepts they still struggled with.

To fix this, I chose a modern web stack that could support real‑time interaction and reliable background processing. Next.js 15 provided server‑side rendering and API routes, which I paired with Supabase for a managed PostgreSQL store and auth layer. LangChain.js gave me a flexible way to chain prompts to Groq’s open‑source‑compatible 120‑b model, balancing the need for powerful language understanding with predictable latency. I used Inngest to schedule interview sessions and rubric evaluations after each learning sprint, ensuring that the system could re‑engage learners without manual intervention. Serwist powered the progressive web app capabilities, so users could review interview feedback even when offline.

The workflow begins with the AI drafting a learning track that the learner immediately owns and edits. The learner can then explore any topic in depth through a chat interface. Once they feel ready, an AI interviewer asks targeted questions and scores the answers against a rubric that was authored alongside the original prompt. If a learner skips a concept or performs poorly, Inngest queues a reminder that resurfaces the gap at an appropriate interval.

The result is a learning experience that only progresses when the learner proves understanding. Gaps are systematically identified and revisited, turning a one‑off content generation session into a continuous mastery loop.