Concourse Academy: Evidence Changes the Path

Build real expertise through meaningful work, explanation, and an evidence-driven path that changes as you do.

Comment

Concourse Academy is an AI-guided environment for developing deep expertise—not completing a finite course.

Most education products move every learner through the same linear sequence toward the same endpoint. Real expertise is less orderly. People begin with different capabilities, develop through meaningful work, follow different fascinations, and sometimes need help from someone who has already mastered the exact thing blocking them.

Our vision is a living territory of knowledge and skills that a learner can navigate throughout their development. Concourse helps them:

  • locate what they can genuinely demonstrate today;
  • see how capabilities connect across a governed domain roadmap;
  • choose the next move most likely to produce capability growth—not engagement;
  • learn through contextual missions shaped by current ability and personal fascination;
  • explain, apply, and transfer ideas instead of merely consuming content;
  • build a portable, inspectable record of evidence that can be questioned and revised;
  • eventually consult learners or vetted experts who already own the capability at hand.

The working Build Week prototype proves the core loop:

  1. A conversational discovery learns Maya's goals, experience, preferences, and fascinations without changing the governed domain itself.
  2. A practical diagnostic locates her starting frontier on an Applied Machine Learning capability map.
  3. Concourse ranks possible next moves and compiles a mission for her current evidence and context.
  4. Maya repairs a flawed football shot-prediction evaluation in a versioned notebook, then explains the result in her own words using the Feynman method.
  5. Concourse assesses the frozen work and confirmed explanation against explicit criteria. Deterministic policy—not the model—decides whether the evidence supports a scoped capability update.
  6. The evidence, assessment, recommendation, and state change remain inspectable and challengeable in her learning record.

How AI is used: OpenAI models make discovery and learning material conversational and adaptive, while governed schemas, bounded repair, traceable provenance, and deterministic fallbacks constrain their authority. AI can propose; it cannot silently award mastery or rewrite the learner's history.

How we built it: Codex was our primary engineering collaborator throughout Build Week. We used it to challenge the thesis, develop the specification, implement and refactor the full-stack product, build the evidence ledger and bounded notebook runner, write extensive tests, conduct browser-led UX audits, and automate the submission video. GPT-5.6 provided critical product, architecture, UX, and narrative review; we turned its findings into small testable changes instead of adopting them automatically.

Current boundary: this is a no-auth, single-learner Applied Machine Learning prototype using synthetic data. It implements one complete evidence-driven journey. Multi-domain roadmap authoring, live human matching, and durable portable credentials are the next horizon—not claims about the submitted build.

Built With

Share this project:

Updates