Interview Prep Tracker · Jiaxiang Tang

Week 1

Map the interview surface and rebuild LLM foundations

0/0 done

Goal: turn scattered knowledge into a coherent map of what interviews can test.

Study blocks

Watch the first CS336 Language Modeling from Scratch lectures and write a one-page topic map.
Review transformer basics with CS224N self-attention notes and The Illustrated GPT-2: tokenization, embeddings, attention, residuals, layer norm, MLP, and causal masking.
Create a local PyTorch interview notebook or script template using the PyTorch docs for tests, timing, and shape-printing helpers.
Solve 6 baseline coding problems from LeetCode 75 or NeetCode Blind 75: arrays, hash maps, two pointers, stack, BFS, and binary search.

Deliverables

Make a ranked list of weak topics across ML coding, general coding, math, research pitch, and behavioral stories using Alisa's interview-type breakdown as the checklist.
Start a company tracker with role, recruiter, interview format, team interest, deadline, and next action.
Do one 45-minute coding session with autocomplete and AI assistance fully disabled.

Week 2

Math, backprop, and general coding foundations

0/0 done

Goal: refresh the concepts that silently appear inside ML coding and technical discussions.

Study blocks

Review probability with Alisa's math notes: expectation, variance, common distributions, Bayes rule, MLE, and sampling.
Review linear algebra with Alisa's math notes: matrix multiplication, eigenspaces, SVD intuition, projections, norms, and shapes.
Review calculus for ML with CS231N backpropagation notes: gradients, Jacobians, chain rule, softmax derivative, and cross-entropy gradients.
Solve 8 LeetCode 75 / NeetCode Blind 75 problems focused on recursion, trees, heaps, intervals, and graph traversal.

Deliverables

Implement a tiny neural net forward and backward pass from scratch using NumPy-style operations after reading CS231N Backpropagation.
Write a two-page math cheat sheet for probability, linear algebra, and calculus formulas you forget under pressure.
Do one 60-minute general coding mock and record every hesitation or bug pattern.

Week 3

PyTorch fluency and attention from scratch

0/0 done

Goal: make the most common ML coding interview task feel routine.

Study blocks

Drill tensor shape manipulation with the PyTorch tensor docs: view, reshape, transpose, einsum, matmul, gather, scatter, and masking.
Implement scaled dot-product attention with causal masks and padding masks from memory using the CS224N transformer note as reference.
Implement multi-head attention, including projection shapes and output projection; compare against torch.nn.MultiheadAttention.
Implement a minimal training loop with optimizer, scheduler placeholder, gradient clipping, eval mode, and checkpoint stub using the PyTorch optimization tutorial.

Deliverables

Build a transformer block from scratch using CS336 Assignment 1 and test shape, mask behavior, and overfit-on-tiny-batch behavior.
Create a personal debugging checklist for off-by-one errors, shape mismatch, device mismatch, dtype mismatch, and masking bugs.
Do one ML coding mock focused on implementing attention or a transformer component.

Week 4

Inference, decoding, and transformer muscle memory

0/0 done

Goal: connect implementation skill with LLM systems questions that show up in research and MTS interviews.

Study blocks

Implement greedy decoding, temperature sampling, top-k sampling, and nucleus sampling with the Hugging Face generation strategies guide.
Implement beam search and write down failure modes, length penalty issues, and complexity using the generation strategies guide.
Study KV cache mechanics, batching, prefill vs. decode, and why memory dominates serving with the vLLM / PagedAttention paper.
Review vLLM-style serving concepts in the vLLM docs: paged attention, continuous batching, scheduling, and SLO tradeoffs.

Deliverables

Re-implement a tiny decoder-only transformer from a blank file in under 90 minutes, then compare against CS336 Assignment 1.
Write a one-page inference systems sheet using vLLM docs and the JAX scaling book: throughput, latency, batching, KV cache, parallelism, and memory bottlenecks.
Do one technical discussion mock on inference system design and record gaps.

Week 5

Research pitch and deep technical discussion

0/0 done

Goal: explain your research judgment clearly and defend design choices under pressure.

Study blocks

For each major project, write the problem, motivation, technical challenge, method, result, limitation, and next step using Alisa's research-discussion advice.
Prepare 30-second, 3-minute, and 15-minute versions of your main research pitch, guided by the preparation section.
Practice experiment-design prompts: ablations, baselines, metrics, negative results, scaling, and follow-up studies.
Review every paper and project on your CV and prepare a concise answer for why it matters.

Deliverables

Build a job-talk outline centered on one coherent research direction, not a disconnected project list; use the job-talk notes as the target format.
Create role-specific pitch variants for LLM systems, post-training, distributed systems, and AI infrastructure roles.
Do one research discussion mock and ask the mock interviewer to press on weak assumptions.

Week 6

Breadth questions and company-specific cramming

0/0 done

Goal: prepare for rapid-fire technical breadth and tailor your study to upcoming interview scopes.

Study blocks

Build a rapid-fire sheet using Alisa's LLM notes for positional encodings, attention variants, distributed parallelism, RLHF, PPO, GRPO, evaluation, and safety.
Review distributed training with the JAX scaling book: data, tensor, pipeline, sequence, expert parallelism, ZeRO, all-reduce, and communication bottlenecks.
Review post-training with Policy Gradient for LMs, OpenAI Spinning Up PPO, and the GRPO guide: SFT, preference data, reward models, DPO, PPO, GRPO, reward hacking, and evaluation.
For each active company, infer the interview scope from recruiter notes, team work, role description, and recent papers or engineering posts.

Deliverables

Make 40 short-answer flashcards for breadth topics and run two timed review passes.
Write one-page company briefs for your top 5 targets: team fit, likely interview topics, questions to ask, and tailored pitch keywords.
Do one rapid-fire technical mock and mark every answer as strong, partial, or weak.

Week 7

Behavioral stories, interview sequencing, and stamina

0/0 done

Goal: prevent blanking on simple behavioral questions and avoid burning out before high-priority interviews.

Study blocks

Create a behavioral story bank for conflict, failure, leadership, ambiguity, collaboration, technical disagreement, mentoring, and ethical judgment using the behavioral interview section.
Convert each story into STAR format with a direct answer to the question, not just a narrative.
Prepare answers for AI safety, societal impact, responsible deployment, and tradeoffs in model release using NIST AI RMF as a framing reference.
Sequence companies by priority, readiness, recruiter pressure, headcount uncertainty, and expected timeline using Alisa's timeline advice.

Deliverables

Do one behavioral mock and require the interviewer to interrupt when the answer does not answer the question.
Pick 1-2 lower-stakes interviews or mocks for practice before top-choice final rounds.
Block sleep, exercise, and no-interview recovery windows on the calendar to protect interview stamina.

Week 8

Full simulation, final polish, and negotiation prep

0/0 done

Goal: simulate real interview pressure and prepare for the offer stage before it arrives.

Study blocks

Complete one ML coding mock with a blank editor, timer, and no AI assistance; use CS336 Assignment 1 as the drill bank.
Complete one general coding mock from LeetCode 75 or NeetCode Blind 75, focused on clean communication and edge cases.
Complete one technical discussion mock and one research pitch mock.
Review all mock notes and convert recurring mistakes into a final drill list.

Deliverables

Create a negotiation sheet with target package, minimum acceptable package, priorities, competing processes, and information you will not disclose using the negotiation section.
Write recruiter-call scripts for timeline extension, competing offers, compensation ask, team-fit questions, and declining politely using Alisa's recruiter-call advice.
Define the day-before-interview routine from the day-of-interview advice: light review, no late cramming, materials ready, and sleep protected.

8-week interview preparation tracker

Weekly Focus

Core rule

Target roles

Weekly rhythm

Detailed Plan

Map the interview surface and rebuild LLM foundations

Study blocks

Deliverables

Math, backprop, and general coding foundations

Study blocks

Deliverables

PyTorch fluency and attention from scratch

Study blocks

Deliverables

Inference, decoding, and transformer muscle memory

Study blocks

Deliverables

Research pitch and deep technical discussion

Study blocks

Deliverables

Breadth questions and company-specific cramming

Study blocks

Deliverables

Behavioral stories, interview sequencing, and stamina

Study blocks

Deliverables

Full simulation, final polish, and negotiation prep

Study blocks

Deliverables

Interview Week Routine

Two days before

Day before

Immediately after