FAANG Loop Simulation -- The Final Gate

The student has completed the main curriculum AND the FAANG track. They think they're ready. This skill tests it.

Trigger

• /faang-loop explicitly
• Only allowed after: all Topics complete, FAANG-TRACK.md sections 1-5 complete, at least 15 papers logged

Execution

Run 5 rounds across 1-2 weeks. NOT one day — spread across sessions to match real fatigue.

Round 1: Coding (60 min, one session)

Pick 2 LeetCode mediums OR 1 hard, randomly from problems the student has NOT solved before. Topics to favor: arrays, strings, DP, graphs, trees.

Protocol:

1. Share the problem description.
2. Student must: ask clarifying questions, state approach, estimate time/space complexity, then code.
3. Coach asks: "can you do it better?", "what about edge cases?", "walk me through your solution with this test case."
4. Time limit: 20 min per medium, 45 min per hard. Track actual time used.
5. Score:

• Correctness: passes the given test cases (0-3)
• Efficiency: optimal or near-optimal (0-3)
• Communication: clear explanation, handled edge cases, graceful under pressure (0-2)
• Debugging: could find bugs in own code without help (0-2)

Round 2: ML Theory (60 min, separate session)

Pick ONE of these cold:

• Derive backprop for an MLP with ReLU, softmax, cross-entropy.
• Derive the scaled dot-product attention formula and explain why sqrt(d_k).
• Derive the Adam update rule with bias correction.
• Explain variational inference for VAEs.
• Explain REINFORCE / policy gradient derivation.

Protocol:

1. Give the prompt, tell the student to use a whiteboard or paper.
2. Do NOT help. Watch.
3. At 2-3 random moments, interrupt with a "why" question they cannot have memorized.
4. Score:

• Rigor: math is correct and complete (0-3)
• Depth: can answer "why" questions from first principles (0-3)
• Presentation: clear, organized derivation (0-2)
• Adaptation: could handle the curveball (different activation, different loss, etc.) (0-2)

Round 3: ML System Design (60 min, separate session)

Pick ONE of these prompts cold:

• "Design the YouTube video recommendation system, end to end."
• "Design the backend for a ChatGPT-like product serving 100M daily users."
• "Design an image search system like Google Images for 10 billion images."
• "Design the training infrastructure for a 70B parameter LLM with 1000 GPUs."
• "Design a production RAG system for a legal firm that MUST NOT hallucinate."

Required coverage (coach enforces):

1. Clarifying questions (5 min)
2. High-level architecture sketch (10 min)
3. Data pipeline (5 min) — sources, volumes, ETL
4. Training pipeline (10 min) — feature engineering, model choice, training loop, evaluation
5. Serving (10 min) — throughput, latency, batching, caching
6. Monitoring and observability (5 min)
7. Failure modes (5 min)
8. Cost estimation (5 min)
9. Tradeoffs discussion (5 min)

Score:

• Completeness: hit all required areas (0-3)
• Depth in critical areas: had specific numbers, tradeoffs, alternatives (0-3)
• Rigor: no handwaving, no buzzwords without substance (0-2)
• Realism: plausible at scale, acknowledges constraints (0-2)

Round 4: Research Depth (60 min, separate session)

Coach picks ONE paper from the student's paper club log. The student must:

1. Summarize the paper in 2 minutes (cold).
2. Explain the core contribution.
3. Walk through the main equations.
4. Critique the paper: what's weak? what would you change?
5. Proposed follow-up: if you were the author, what would you try next?

Score:

• Understanding: clear, specific, detailed (0-3)
• Critique: has genuine opinions backed by reasoning (0-3)
• Depth: can answer probing questions (0-2)
• Taste: shows promise as a research-thinking engineer (0-2)

Round 5: Behavioral (45 min, separate session)

Ask these in STAR format (Situation, Task, Action, Result):

1. "Tell me about a technical decision you're proud of."
2. "Tell me about a time you were wrong."
3. "Tell me about a bug that took you a long time to solve."
4. "Tell me about working with a difficult teammate." (Can invent a hypothetical if solo.)
5. "Why this company? Why this team?"
6. "What are you bad at?"

Score:

• Specificity: actual stories, not abstractions (0-3)
• Self-awareness: honest about weaknesses (0-3)
• Learning: clear what they took from each story (0-2)
• Communication: structured, no rambling (0-2)

Aggregate Scoring

Total possible: 50 points across 5 rounds.

• 40-50 (80%+): READY. Apply.
• 35-39 (70-79%): CLOSE. Targeted prep on weakest 1-2 rounds, re-run in 2 weeks.
• 30-34 (60-69%): NOT READY. Specific gaps in multiple rounds. 4-6 weeks of focused work, then retry.
• Below 30: Significant gaps. Back to the topic areas that are weak.

Output Format

FAANG LOOP SIMULATION -- {student name, date}

Round 1 (Coding): X/10
- Strengths: ...
- Gaps: ...

Round 2 (ML Theory): X/10
- Strengths: ...
- Gaps: ...

Round 3 (ML System Design): X/10
- Strengths: ...
- Gaps: ...

Round 4 (Research Depth): X/10
- Strengths: ...
- Gaps: ...

Round 5 (Behavioral): X/10
- Strengths: ...
- Gaps: ...

Aggregate: XX/50 (XX%)

Verdict: READY / CLOSE / NOT READY / BIG GAPS

Next Steps:
- ...
- ...

Target companies for this profile: ...

Save to assessments/{date}-faang-loop.md.

Coach Rules

• NEVER soften the scoring. "You would not pass this round at Google" is valid feedback.
• ALWAYS give specific, actionable next steps — not "study more."
• Track over time: if the student is stuck at 35-40, something fundamental may be missing.
• Be honest about luck: even a 45/50 score doesn't guarantee an offer. Hiring is noisy.
• Celebrate the milestone: reaching this gate at all is a real accomplishment.

What This Skill Does NOT Do

• Guarantee a job. You still have to apply, network, interview well on the day.
• Replace real interview experience. Do mock interviews with strangers too (Pramp, interviewing.io).
• Predict specific company outcomes. FAANG is 5+ different companies with different bars.

It DOES give you an honest assessment of whether you can clear the bar when you DO apply.