IT Talent Skill Framework — Software · Hardware · AI

Skill standards for selecting students into the Special IT Faculty — software, hardware, AI — based on the criteria of Stanford, CMU, and Tsinghua.

1. The Skill Graph

Six reference universities, each selecting for its own independent group of the skills (Stanford · CMU · Tsinghua · Cambridge · Oxford · Berkeley) — so the radars have different axes, not one shared shape. Toggle two or more to overlay them on one graph for comparison (a skill not in a style sits at the center). Team leads and board mark the styles they ❤ love.

2. The Skills, Explained — In Detail

graphed = scored by test problems, becomes a radar axis. qualitative = assessed from documents and interview, never scored. A skill is approved when more than 2 of the 4 members accept it (3 or 4 of: 3 team leads + board) — then it moves to the test list.

Intellectual Vitality umbrella

Being deeply curious and loving learning because you genuinely enjoy understanding things — not just to get grades, win awards, or make money. Stanford rates this separately because it wants students who will create new knowledge, not just consume it. It is the umbrella over the four skills below: curiosity, joy in learning, persistence, and independent learning.

Strong signals
  • Reading science articles or books far beyond the school program — for fun, not for credit.
  • Building your own small app, game, or gadget just to understand how it works inside.
  • Spending a weekend on a question nobody assigned — "how does a search engine decide what comes first?"
  • Asking deep questions and exploring beyond what is required.

Genuine Curiosity graphed

A natural desire to understand how things work — refusing to accept facts without asking why.

Strong signals
  • Instead of just accepting "AI can recognize faces," you keep asking: How does it actually do that? Why does it need thousands of example photos to learn? When does it make mistakes? Could it be fooled — and how would we test that?
Measured by: English test "write 3 questions about this passage" (scored on question quality); question-asking behavior in the verbal interview.

Joy in Learning for Its Own Sake qualitative

Learning because learning itself is rewarding.

Strong signals
  • Learning a new programming language, instrument, or foreign language nobody asked you to learn.
  • Watching a documentary about how the internet works simply because you find it interesting.
Not this
  • "I only learn this because it will be on the exam."
Assessed in: warm-up interview ("what did you last learn that nobody asked you to?").

Persistence / Resilience graphed

Continuing despite setbacks and difficulties. In the famous West Point study, grit predicted completing elite training better than the official admission score — and CMU's review committees name "persistence and resilience in the face of challenge and adversity" as an explicitly valued quality.

Strong signals
  • Spending weeks on a bug in your project that everyone else gave up on — and fixing it.
  • Entering the competition again after losing, with a better plan each time.
  • Continuing a long project after many failed attempts — changing the approach instead of quitting.
Measured by: long multi-step math problem; programming debugging problem that yields to systematic search, not one insight; "hardest bug" interview question.

Independent Learning / Teachability graphed

How fast you absorb a new concept and correctly apply it — the criterion elite universities probe hardest in interviews, and the #1 trait in hired-without-a-degree stories at top technology companies.

Strong signals
  • A test problem teaches you an unfamiliar notation, and two lines later you use it correctly.
  • Given a hint in the interview, you absorb it and immediately build on it.
Measured by: "learn-then-apply" problems in the math and programming tests; reaction to hints in the verbal interview.

Mathematical Reasoning graphed

Rigorous step-by-step analytical thinking — ranked the #1 skill by employers worldwide and the foundation of every machine-learning course.

Strong signals
  • Proving a pattern instead of stating it (e.g., why perfect-square lockers stay open).
  • Translating a formula into a working procedure.
Measured by: math test proof problems; complexity-analysis programming problem; Track-D interview questions.

Creativity / Risk-Taking graphed

Finding the elegant, non-obvious path — and being willing to try an approach that might fail. CMU looks for it as independent problem-solving shown in real projects; Tsinghua calls it innovative thinking.

Strong signals
  • Spotting the shortcut where everyone else grinds through the slow standard path.
  • Proposing an unusual design in an open-ended problem and defending its trade-offs.
Measured by: math problem with a slow path and an elegant shortcut (shortcut rewarded); open-ended algorithm-design problem; design questions in the interview.

Communication & Persuasion graphed

Explaining your reasoning so a teammate can follow it — and arguing a position convincingly. In a competitive research world, an unexplained result is an unused result.

Strong signals
  • A pseudocode answer whose logic a non-author can re-implement.
  • A persuasive professional email that proposes a solution, not just an excuse.
Measured by: clarity scores on algorithm-only answers (problems 7–9); English persuasive writing; presentation segment of the group stage.

Collaboration & Initiative graphed

Working well with others — and starting things without waiting to be told. CMU's review committees explicitly value leadership and concern for community alongside technical depth.

Strong signals
  • Building on a teammate's idea instead of competing with it; bringing quiet members into the discussion.
  • Volunteering structure ("let's list the options first") when a group is stuck.
Measured by: observer tally sheet during the group discussion; directed Q&A in presentations.

Integrity qualitative

Being honest, ethical, and trustworthy.

Strong signals
  • Reporting experimental results accurately, even when they contradict your hypothesis.
  • Not taking credit for others' work; following research ethics.
  • Saying "yes, I've seen this problem before" when offered a fresh one in the interview.
Assessed by: the "have you seen this problem?" honesty check; consistency between declared background and demonstrated skill. Violations are eligibility issues, not point deductions.

Depth of Commitment qualitative

Sustained dedication to one meaningful interest over years. Committees prefer one deep passion over many disconnected activities, because depth predicts significant contribution.

Strong example
  • 4 years deepening one passion: simple games → a website for the school club → the database behind it → a small published app — every step deeper in the same direction.
Weak example ("well-rounded but not distinguished")
  • Robotics 2 months · blockchain 1 month · stock trading 3 weeks · biology 2 months — 20 clubs, 10 hackathons, 15 online courses, but no substantial project or long-term mastery.
Assessed from: application documents, portfolio history, warm-up interview.

Leadership graphed research

Influencing and inspiring peers, starting projects, and giving a group direction — distinct from simply collaborating. Named explicitly by CMU (a core pillar: "produce the next generation's leaders"), MIT, Berkeley, and Tsinghua.

Strong signals
  • Founded or ran a club, team, or project — moved others toward a goal.
  • In the group stage, raises the whole team's output, not just their own.
Measured by: group discussion observer notes; "tell me about something you organized" interview question; portfolio.

Attention to Detail / Rigor graphed research

Precision and correctness — checking edge cases, testing assumptions, and intellectual honesty under scrutiny. Oxford and Cambridge select for mathematical rigor; DeepMind and Anthropic name rigor and "intellectual honesty under scrutiny" as what distinguishes strong researchers.

Strong signals
  • Spots the empty-list / off-by-one / overflow case before being told.
  • Reports a result honestly even when it contradicts their hypothesis.
Measured by: bug-finding & edge-case credit on programming problems; proof completeness in math; honesty check in the interview.

Ownership / Accountability graphed research

Sees work through end-to-end and cares whether the whole thing ships — not just "my part works." Remote-first companies hire "managers of one"; CMU values students who initiate with a sense of purpose.

Strong signals
  • When blocked, finds a way forward instead of waiting to be told.
  • Follows a task to completion and keeps stakeholders informed.
Measured by: the professional-email English problem; "time you faced a blocker" interview question; final project.

Systems Thinking graphed research

Reasoning about whole systems and how their parts interact — trade-offs, data flow, failure modes — not just a single function. DeepMind explicitly names "systems-level thinking" as a core researcher quality.

Strong signals
  • Asks where the data comes from and what could go wrong across the whole pipeline.
  • Weighs trade-offs (speed vs. memory vs. accuracy) rather than optimizing one part blindly.
Measured by: the "design an AI service" group task; complexity/trade-off discussion in the verbal stage.

3. Career Target Profiles — What Are We Selecting For?

Dashed purple = the ideal shape each career demands; solid = closest reference profile. The gap between them is what mentoring must close.

4. How We Pick the Top 10

The rule we apply, in order:

5. Summary — Total Skill Status

All scores, votes, and decisions in one live table. Green = strongest axis, red = weakest.

6. Team Lead Requirements — Editable

Each team lead edits their team's required shape: slide levels · “del?” marks a shared skill for deletion (kept in the database, removed later at board review) · quick add an existing skill (☆) · or define a new skill (★) in full card format. Saves automatically.

7. Our Own Skill Framework

The final selected skill set — every skill the committee approved (3+ of 4) plus every skill the team leads added (auto-approved), in one list, with how strongly each is tested. This is the framework we build our selection on.