Full-Stack & Applied Data Science
A study guide for senior/staff Data Scientist loops where the work is shipping production models end-to-end — calibrated against a Staff Data Scientist role at a fraud/identity platform and a Data Scientist role at a multimodal-sensor AI company. 17 interactive HTML chapters with sticky navigation, syntax-highlighted code, copy buttons, drill modes, and dark mode.
If you only have an hour
Open 00-START-HERE first. Then drill 15-interview-questions with drill mode on. Reread 16-day-of the morning of.
Drill: Interview Questions
Hide-show answers · per-Q tracker
Drill: Coding Problems
Python · multiple approaches
Core: Production ML
Code, tests, monitoring, drift
Section A · Orient
00
Start Here
Master index, study schedule, the reframe that separates "ML researcher" from "full-stack DS who ships."
01
The Roles Decoded
Staff Data Scientist (fraud / identity) and Data Scientist (multimodal sensor AI), side by side. What each team actually does, what the stack means, what the loop is really testing.
02
Positioning From Scratch
Mindset before content — how to interview honestly when you've done some production ML but not at staff scale, and how to lean on adjacent experience without overclaiming.
Section B · Core Data Science
03
ML Fundamentals
Supervised models for tabular data — logistic regression, gradient boosting, calibration, evaluation under skew, model selection at staff bar.
04
Feature Engineering
"Inventive feature engineering" demystified. Entity-level features, leakage avoidance, drift, feature stores, and the hard part: features that encode domain insight.
05
Fraud & Imbalanced Data
Class imbalance, cost-sensitive learning, precision-recall tradeoffs, threshold tuning at real-world fraud rates, evaluation when labels are scarce or delayed. Fraud-domain-flavored.
06
Prompt Engineering, Applied
N-shot examples, prompt templates, lens/parameter configuration, prompt-as-config patterns, iterating prompts against an eval set. Solutions-Engineering-flavored.
07
Time-Series & Signals
Signal processing basics, time-series features, sensor data prep, alignment and resampling, video and multimodal data prep. The "raw data to testable results" workflow.
08
Data Pipelines, Applied
Data acquisition, labeling workflows (Encord and others), iterative preprocessing cycles, feature pipelines, training data versioning, the loop from raw asset to model-ready dataset.
09
Production ML
Writing production code & tests, real-time inference, monitoring, retraining cadence, A/B rollouts for models, on-call considerations. The Staff-DS "end-to-end ownership" bar.
Section C · Coding
10
Coding Fundamentals
Python patterns full-stack DS interviewers reach for — clean classes, generators, testability, vectorization vs explicit loops, useful Big-O intuition.
11
Coding Problems
Drillable problems — feature engineering puzzles, imbalanced-data evaluation, ranking and aggregation, a couple of classic algos. Multiple approaches per problem; drill mode hides solutions.
Section D · Production / Cloud
12
AWS Data Stack
EC2, S3, RDS, Redshift, plus PostgreSQL as the system of record. What the standard fraud-stack implies about how their work is structured, and how to discuss it credibly.
13
MLOps Applied
CI/CD for models, experiment tracking, shadow deploys, canary rollouts, drift monitoring, retraining triggers — the operational glue that separates a Jupyter result from a product.
Section E · Reference & Execution
14
Domain Context
Two domains, one chapter — fraud & identity verification and multimodal sensor-fusion AI. The vocabulary and stakes you need to sound credible.
15
Practice Interview Questions
~30 Q&A across ML, features, evaluation, production, prompts, signals, and domain. Drill mode hides answers; per-question practice tracker; progress saved locally.
16
Day-Of Tactics
Structural moves, traps to watch, recovery patterns, the right questions to ask them, closing statement. Reread morning of.
Study Paths
If you have 7+ days
- Day 1: 01, 02 (orient) → 03 (ML fundamentals)
- Day 2: 04 (features), 05 (fraud/imbalanced)
- Day 3: 06 (prompts), 07 (time-series & signals)
- Day 4: 08 (pipelines), 09 (production ML)
- Day 5: 10, 11 (coding drills on a timer)
- Day 6: 12 (AWS), 13 (MLOps), 14 (domain context)
- Day 7: Drill 15. Reread 16. Sleep.
If you have 2–3 days
01, 02, 03 (ML), 04 (features), the role-relevant 05 or 06, 09 (production ML), 11 (drill 4–5 problems), 15 (drill), 16. Skim everything else.
If you have < 24 hours
01, 02, 04 (features — central to both roles), 05 or 06 (whichever applies), 09 (production ML), 15 (drill all questions), 16. Skim 03, 07, 13 headings only.