CompetenCY

Capture learning as it happens. CompetenCY uses AI to assess collaboration, critical thinking, and other competencies from authentic classroom interactions without disrupting instruction.

Comment

About the Project

Inspiration

Traditional classroom assessment often captures only the final product—a quiz score, an assignment, or a presentation grade. As educators, we wanted to recognize something equally important: the learning process itself.

Students demonstrate communication, collaboration, critical thinking, and other essential competencies through authentic classroom interactions every day. Yet teachers rarely have enough time to observe every learner, provide individualized feedback, and document evidence consistently.

What if every classroom interaction could become meaningful evidence of learning?

CompetenCY was built to answer that question. Our goal is to help educators capture authentic student performance, generate evidence-based feedback, and keep teachers—not AI—in control of every assessment.

How We Built It

CompetenCY is built with React, TypeScript, and Supabase, which provide authentication, PostgreSQL storage, row-level security, private file storage, and serverless backend functions.

At the core of the platform is an AI-powered assessment pipeline powered by OpenAI GPT-5.6 Terra and the OpenAI Responses API.

The workflow includes:

  • AI-assisted assignment generation
  • AI-generated rubrics from learning objectives and competency frameworks
  • Rubric standardization while preserving institutional intent
  • OCR and text extraction from PDFs and images
  • Speech transcription and speaker diarization using ElevenLabs
  • Evidence extraction from transcripts and documents
  • Rubric-aligned competency scoring
  • Timestamped, evidence-based feedback generation
  • Instructor review, editing, and approval before releasing feedback to students

Instead of asking a language model to evaluate an entire transcript at once, we designed an evidence-first evaluation pipeline. Every rubric judgment must reference specific supporting evidence from a student's recording or submission, allowing instructors to verify exactly why feedback was generated.

To improve transparency and reproducibility, every evaluation stores rubric versions, prompt versions, model metadata, and evaluation provenance.

Challenges We Faced

Building an educational assessment platform required solving problems beyond simply integrating an LLM.

Grounding AI feedback in evidence

Early prototypes sometimes attached broad or loosely related transcript passages to rubric criteria. We redesigned the pipeline so every feedback item references validated evidence anchored to specific transcript timestamps or document excerpts.

Supporting multimodal assessment

Students submit audio recordings, PDFs, images, and written responses. We developed a unified evidence representation that allows different modalities to be evaluated consistently while preserving their original sources.

Preserving institutional rubrics

Schools use diverse rubrics with different scales, terminology, and structures. Rather than forcing everything into a single scoring system, CompetenCY preserves original rubrics while allowing optional AI-assisted standardization that instructors can review before use.

Keeping humans in control

The most important design decision was recognizing that AI should assist—not replace—educators. Every report can be reviewed, edited, re-evaluated, and approved by instructors before students receive feedback.

Building reliable AI infrastructure

Educational assessments cannot fail silently. We built an asynchronous evaluation pipeline with retries, provenance tracking, secure institution-scoped access, and validation layers to ensure reliable processing and protect educational data.

What We Learned

Building CompetenCY reinforced several lessons:

  • Educational AI should explain why a score was produced—not simply generate one.
  • Evidence quality is just as important as model quality.
  • Structured outputs improve reliability but still require backend validation.
  • Human review should be part of the product architecture from the beginning, not added later.
  • Assessment is more valuable when it captures learner growth over time instead of isolated scores.
  • AI is most effective when it amplifies teacher expertise rather than automating consequential educational decisions.

Most importantly, we learned that AI can make authentic classroom learning visible at a scale that was previously impossible—while keeping educators at the center of every important decision.

Built With

  • OpenAI GPT-5.6 Terra
  • OpenAI Responses API
  • React
  • TypeScript
  • Vite
  • Supabase
  • PostgreSQL
  • Supabase Edge Functions
  • Deno
  • ElevenLabs
  • Tailwind CSS
  • shadcn/ui
  • TanStack Query
  • React Router
  • Zod
  • Recharts
  • PDF.js
  • Mammoth.js
  • GitHub
  • Vercel

Try It Out

🌐 Live Demo: https://www.competen-cy.com/

💻 Source Code: https://github.com/eko09/practice-patient-pal

Note: The live demo is publicly accessible. The GitHub repository is currently private and should be made public (or shared with judges) before submission.

Built With

Share this project:

Updates