ExaMinds
Junior engineers and technical employees often know the theory but struggle to apply it when real conditions change. Fixed courses and quizzes give limited opportunities to practice consequences safely. ExaMinds turns knowledge into personalized decision games. Two initial assessments identify knowledge gaps and explanation preferences. The learner then enters a physics-driven drilling game, controls pump power and mud density, and maintains pressure within changing limits to avoid a kick or fluid loss. Our team combines applied data science, mathematics and physics teaching, educational course creation, and BP engineering experience. The full local platform also includes Bayesian skill tracking, targeted follow-up scenarios and an independent transfer assessment.
Preliminary score
- Value for the user18.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI22.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing15.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility8.0 / 15
Data requirements, running costs and a clear next step.
- Originality7.0 / 10
A meaningful difference from common solutions.
What the judges wrote
- Claude Opus 5.5Preliminary67/100
- Strengths
- Your drilling simulator gives a concrete, safety-critical use case. Learners control pump power and mud density against a moving pressure window, and the kicks and losses come from a deterministic SI-unit physics engine rather than from the language model. The adaptive layer is real and inspectable: Bayesian Knowledge Tracing that weights hinted answers less, contextual Thompson sampling for choosing explanation formats, and a transparent mission planner. Your testing write-up is unusually honest and specific: 75 automated checks, physics property sweeps across 120 perturbed wells, named failures you fixed, and clear statements that Claude was tested only with mocks and that learning gains are unvalidated.
- To improve
- The public demo runs on prepared content and the Claude path was never exercised live. Show at least a few real generated chapters, with validation pass and fail rates, so the generative AI's contribution is demonstrated rather than optional. Run even a small pilot against a fixed-course baseline, for example five to ten colleagues comparing transfer-exam scores or time to mastery. That would give you the adoption metric the Enterprise track asks for. Add running costs (LLM tokens per learner or chapter, hosting) and a concrete next step with a named pilot partner, and replace the 'Game footage — to be added' slide with real gameplay.
- GPT-6 AstraPreliminary74/100
- Strengths
- Your prototype connects learner diagnostics to adaptive drilling missions with physically computed consequences, rather than stopping at generated lessons. Your separation of AI-authored narrative from deterministic physics and approved assessment keys is a strong safeguard. Your documented tests address concrete failures, and you clearly distinguish implemented adaptation, mocked provider calls and learning outcomes that remain unvalidated.
- To improve
- Your next validation should compare adaptive practice with a fixed-course baseline using unseen transfer tasks, delayed retention and time to demonstrated competence. Demonstrate a live model-backed session, and evaluate learner-model calibration and explanation selection with real learners rather than synthetic sequences alone. Your deployment plan should quantify per-learner operating costs and define expert review, access controls and a small employer pilot before treating skill estimates as proven competence.
The team
Murad Nuraliyev
data
Chingiz KhankishiyevLead
data
Elnur Hagverdiyev
data