🔥 Hardcore Edition · 16-Week War Plan · No Mercy

Data Analyst → Core Data Scientist

A maximum-intensity, quota-driven 4-month regimen for heavy ML / Applied-Scientist loops at Meta, Apple & peers. Built for a mid Python dev with shipped ML projects (CPC prediction, etc.). Every week has hard volume quotas and a mastery gate you must pass — out loud, on camera, no notes — before you advance. Comfort is not on the menu.

4 hrs/day · 7 days/week
3h Technical + 1h Speaking
16 weeks · 4 phases · 16 gates
~1,800 ML problems · 112 recordings

! The Brutal Truth (read before you commit)

Hardcore means hardcore. If you can't hold 4 disciplined hours, 7 days a week, for 16 weeks, stop here and use the lighter plan — half-doing this just burns you out.

On the technical pivot

Your shipped ML projects put you ahead of most analysts. But "I built a CPC model" and "I can derive the loss, implement the optimizer from scratch, design the serving system, and defend all of it under hostile follow-ups" are different leagues. This plan closes that gap by brute-forcing depth + volume + verbal fluency — not by re-teaching basics.

250+ focused hours over 16 weeks is enough to become a competitive junior-to-mid Applied ML candidate. It is not enough for a publications-track Research Scientist. Target correctly.

On speaking — your highest-risk failure mode

You can have flawless math and still fail the loop on delivery. The fluent explainer of a simple idea beats the mute expert every single time. This boot camp is built to make you physically unable to ramble — through daily recorded reps, hard filler limits, and hostile-interrupt simulation. It will feel humiliating in Week 1. That humiliation is the training stimulus.

1 The 3-Hour Technical War Plan ML

Five tested skills: ML coding, ML theory/depth, stats & probability, ML system design, GenAI/LLM systems. Every week carries volume quotas and a gate. Click a header to collapse.

Non-negotiable daily technical quotas

  • 3+ probability/stats problems solved with full working
  • 1 algorithm or component implemented from scratch (NumPy)
  • 10 rapid-fire theory Q&A answered aloud in <60s each
  • 1 concept explained on camera, no notes (feeds Speaking track)
  • Weekly: 1 recorded mastery-gate attempt + 1 timed mini-mock
PHASE 1 · WEEKS 1–4

Foundations under load: Math, Probability & Classical ML you can derive blind

From "I used it" to "I can derive it, code it from scratch, and survive interrogation."
▾

Week 1 Probability & the math under ML

Whiteboard-derive MLE for linear regression, on camera, no notes, in under 6 minutes.

Week 2 Regression from scratch

Code logistic regression + gradient descent from an empty file in <25 min while narrating each line aloud.

Week 3 Evaluation & bias-variance

Given a random business scenario, pick the right metric and defend it against 3 "why not X?" follow-ups, on camera.

Week 4 Tree-based models

Explain "trees vs RF vs boosting, when I'd pick each" in 3 min, then defend it against a skeptic for 5 min.
PHASE 2 · WEEKS 5–8

Deep Learning at depth: Nets, Training, Transformers

From tabular comfort to deep-learning fluency you can whiteboard and code.
▾

Week 5 Neural net fundamentals + backprop

Derive + narrate backprop for a 2-layer net on a whiteboard, no notes, handling one "why?" interrupt mid-derivation.

Week 6 Training deep nets in practice

Deliver a 4-min "my training loss won't drop — here's how I debug it" troubleshooting monologue, structured.

Week 7 CNNs, RNNs, Transformers

Explain self-attention with a tiny worked example (3 tokens) on camera in 4 min.

Week 8 Consolidation + rapid-fire grilling

Survive a 45-min rapid-fire theory mock at ≥80% correct with zero "I don't know" freezes.
PHASE 3 · WEEKS 9–12

ML System Design + GenAI/LLM Systems — your moat

The skill analysts lack entirely. Design production ML & LLM systems end-to-end, out loud.
▾

Week 9 The design framework

Design any system from cold start using the full framework in 35 min, narrated, no template visible.

Week 10 Recommendation & ranking

Defend a ranking-system design through 8 minutes of escalating hostile follow-ups without contradicting yourself.

Week 11 Production reality: scale, drift, monitoring

Narrate the full prod lifecycle of one system (deploy → monitor → drift → retrain → rollback) in 6 min.

Week 12 GenAI / LLM systems

Design and defend an LLM feature's full eval strategy in 8 min, including how you'd catch silent quality regressions.
PHASE 4 · WEEKS 13–16

Conversion: Mock Loops, Portfolio, Interview Fire

Turn "knows the material" into "passes the loop." Pure pressure-testing.
▾

Week 13 ML coding under the clock

Implement k-means from a blank file in <20 min while narrating, then explain its failure modes.

Week 14 Portfolio & project interrogation

Survive a 30-min deep-dive on your CPC project where every choice gets a "why not the alternative?"

Week 15 Full mock loops

Score ≥3/4 on every rubric dimension in a full recorded loop.

Week 16 Sharpen & deploy

Your Week-16 recording is unrecognizable next to Week 1. If not, extend Phase 4 — don't apply yet.

2 The 1-Hour "Unbeatable Speaker" Boot Camp — Hardcore SPEAK

This is not soft-skills practice. It's a daily physical regimen to rewire how you speak under pressure. Every session is recorded. Every session is rewatched. No exceptions for 112 straight days.

Non-negotiable daily speaking quotas & hard standards

  • ≥15 min of recorded impromptu speaking, every day
  • Filler words: Week 1 ≤6/min → Week 16 ≤1/min
  • Zero hedging phrases ("I think maybe sort of", "kind of")
  • Pace: 130–150 WPM, deliberate pauses instead of fillers
  • Every technical answer: answer-first, then exactly 3 points
  • Rewatch every recording once; log 1 fix for tomorrow

The fixed daily 60-minute format (intensity-loaded)

0–10 min · WARM-UP & FILLER KILL
Diction + silence training
2 min tongue-twisters/articulation, then speak 5 min on a random topic. Rule: any "um/uh/like/so" → full stop, restart the sentence. Count fillers. You may not finish until you beat yesterday's rate.
10–25 min · STRUCTURE UNDER LOAD
Pyramid Principle + STAR reps
Draw a random prompt. 15-second silent structuring, then deliver: conclusion first → exactly 3 points → detail. Behavioral prompts use STAR with a quantified result. Do 3 reps back-to-back, no breaks.
25–45 min · EXPLAIN THE HARD THING (your differentiator)
Today's technical topic, three layers, two audiences
Explain today's ML concept (1) intuition in one sentence, (2) the math term-by-term, (3) practical why. Then re-explain it to a non-technical PM. Both on camera, no notes. This is the exact loop skill and where most candidates die.
45–60 min · PRESSURE / HOSTILE SIM
Thinking on your feet under fire
Run one drill from the bank below with a hostile-interviewer setting: random interrupts, "why not X?", skeptical follow-ups. Record, rewatch at 1.5×, log one fix.

Weekly escalation

WeeksSpeaking focus (hardcore)Standard to hit
1–4Kill fillers/hedging; explain equations (MLE, GD, log-loss) cleanly≤4 fillers/min; equation explained in <3 min
5–8Trade-off duels; explain backprop/attention at two audience levels≤2 fillers/min; survive 3 interrupts
9–12Architecture-over-video narration; structured thinking under interruption≤1.5 fillers/min; defend 8-min grilling
13–16Full behavioral STAR bank; whiteboard narration; loop delivery≤1 filler/min; ≥3/4 communication rubric

3 The Hardcore Drill Bank SPEAK

Technical speaking's failure mode: people who can do the math go mute trying to narrate it. Train narration as hard as the math. Rotate these in the 45–60 min block.

1 · Equation narration. Explain today's equation in three layers without freezing. Hard mode: a metronome forces one logical step every 8 seconds — no stalling.
2 · The trade-off duel. Two models (XGBoost vs NN, RAG vs fine-tuning, L1 vs L2). Argue A 90s, B 90s, then a 30s verdict. Hard mode: a coin flip decides which side you defend, announced at the buzzer.
3 · Architecture-over-video. Narrate every box and arrow as if screen-sharing to a remote panel. Hard mode: no drawing allowed — describe the whole system in words only.
4 · The interruption / hostile drill. A phone alarm fires at random 20–40s intervals; each ring = an interviewer cutting in with "why not X?" You must land the sentence, answer, and resume. This is the single most important drill.
5 · Two-audience drill. Explain a concept to an ML scientist, then a non-technical PM, then a hostile VP who thinks ML is hype. Three registers, back-to-back.
6 · Record → transcribe → red-pen. 3×/week, transcribe a recording. Highlight every filler, hedge, and run-on. Hard mode: rewrite the transcript into a tight version, then re-record matching it.
7 · Whiteboard narration. Solve a problem while narrating every step — think and talk simultaneously. Hard mode: a 10-min hard cap; if you go silent >5s, restart.
8 · Cold-open ambush. Random prompt + 5-second timer to structure, then 90s answer-first. Trains structure-under-zero-prep — the exact loop reflex.
9 · Shadowing. Pick a great technical speaker (conference talk). Shadow their delivery line-by-line, matching pacing, pauses, and emphasis. Builds rhythm into muscle memory.

★ Interview Practice Arena LIVE DRILL

Pick a category, hit Draw question, start the timer, and answer out loud on camera — answer-first, then 3 points. Treat the clock as the interviewer's patience. The full bank is browsable below the drill.

Choose a category and draw your first question.
05:00
Questions drawn this session: 0 · Pool size: 0
🎥 Always record. Structure: answer first → exactly 3 points → detail. When the timer hits zero, an interviewer would interrupt — so would you ramble past it? Rewatch once, log one fix.

Browse the full bank

4 The Self-Grading Rubric (score every mock 1–4)

Score yourself honestly on each recorded mock. A "1" anywhere is a stop-and-fix. You do not apply until you hit ≥3 across the board (Week 15 gate).

Dimension1 — Fail2 — Weak3 — Passing4 — Strong
CorrectnessWrong/major gapsRight idea, shaky detailsCorrect, minor slipsCorrect + nuanced
StructureRamblingSome orderAnswer-first, 3 pointsCrisp, signposted
Depth under follow-upFolds at "why?"1 level deep2–3 levelsAnticipates, never folds
Delivery>5 fillers/min, hedging3–5 fillers/min≤2 fillers/min≤1, confident pauses
Composure under fireFreezes/panicsVisibly rattledRecovers cleanlyCalm, even enjoys it

5 How to spend the 4 hours, every single day

3-Hour Technical block

75 min learn/derive → 75 min code or design a problem (hit the daily quota) → 30 min close everything and re-explain today's topic on camera, no notes. The last 30 min is where retention happens — and it's a free speaking rep.

1-Hour Speaking block

The fixed 4-part format (10/15/20/15), always recorded, always rewatched. Today's technical topic is the content you explain. The two tracks compound — that's the whole point.

⏱ Discipline > motivation. 4 focused hours every day for 112 days beats heroic bursts. Protect the streak like your offer depends on it — because it does.

6 Make-or-Break Factors

1 · Depth over breadth — always

Truly owning logistic regression, trees/boosting, and neural nets (math + code + verbal) beats name-dropping 20 algorithms. Interviewers drill down; shallow knowledge dies on the second follow-up.

2 · System design is your moat

The one thing analysts lack entirely and the strongest "more than an analyst" signal. Phases 3's framework must be reflexive. Do not shortchange Weeks 9–12.

3 · Your projects must survive interrogation

Defend every choice in your CPC predictor for 30 minutes. A defended project beats a dozen tutorials.

4 · Delivery is a graded section

The fluent explainer of a simple idea beats the mute expert. Skipping the speaking hour is the fastest way to fail with good knowledge.

5 · Respect the gates

Don't advance a phase until you've passed its mastery gate on camera. The gates are the whole point — they convert "studied" into "can perform."

6 · Plan for month 5

Clearing this bar usually still needs a stretch of pure mock loops to convert "credible" into "offer." Early rejections are paid mocks, not verdicts. Momentum is the real test.

Realistic outcome in 4 hardcore months

✅ A competitive junior-to-mid Applied ML / Core DS candidate who passes theory rapid-fire, survives system design, and presents with genuine confidence.

⚠️ Not a publications-track Research Scientist — different game, longer timeline.

❌ No plan guarantees an offer; the loop and luck aren't yours to control. Control the prep, respect the gates, run the numbers game — and the odds tilt hard your way.

0% complete
0 / 0 tasks