1. Highest-value order
Work top to bottom.
Move on when the exit test passes; do not keep polishing a comfortable topic while a higher-value lane is still weak.
Priority 1 — genuine STAR depth Resume-derived
Open chapter 12’s STAR and why-chain method,
then verify the locked project spread.
Chapter 12 is the complete spoken source, including
all sixteen LP cards,
second genuine stories, and
later-loop prompt mappings.
- Rehearse four leads: Ylogx Dive Deep, Horizon Deliver Results, Argus failed quality bar, and IQVIA technical disagreement.
- For each lead, answer: why this approach, what failed first, how measured, what changes at 10×, and what I owned.
- Failure is Argus at 73% mAP before the 89% gate; do not invent a missed SLA or manager quote.
- Deadline is honest sequencing and early risk communication; do not invent a missed ERC or missed 99.9% target.
- Conflict is hybrid versus vector-only or database RLS versus app-only checks, followed by commit on evidence.
Exit test:
Give one 80-second STAR and survive five consecutive “why?” probes without adding a new metric or fictional event.
Priority 2 — preserve coding posture IE-asked
Start with the chapter 05 Live Code sequence.
For the 30-second “what pattern / why brute dies / TC+SC” scan use
chapter 25, then the named-family card in
chapter 24 (LP scan is last in that same file).
Rehearse Distance K without a parent map,
dynamic k-th largest,
grid count plus max reward, and
maximum-sum switching lists.
- Say clarifications, brute force, optimal invariant, Java, dry run, time, and space.
- Use
long when sums can overflow and safe comparators for heaps.
- Do not stamp Maximum Sum Linked List as LC 962; switch accumulated sums only at common nodes.
- Do not let “LP-only” or “Fluency” recruiter framing remove the coding warm-up.
Exit test:
Write one complete Java solution in 25 minutes and narrate the invariant while typing.
Priority 3 — code-first Fluency IE-asked
Use chapter 07’s slot rule:
the pasted prompt controls the work.
LC 7623949’s Fluency-labeled slot asked
Distance K in O(n) without a parent map.
LC 7643209 was an intern report with an
unnamed min-heap problem;
it was not Distance K and is not an FTE forecast.
tree / graph / heap / stack / DP pasted → clarify and write Java first
class-design prompt → clarify actors and invariants
verbal AI question → project + mechanism + verification + boundary
code complete, Fluency still requested → one grounded AI sentence, then stop
For verbal depth, use
LangGraph mechanism,
Azure Hybrid plus Semantic retrieval, and
where not to use an LLM.
Exit test:
Correctly distinguish “FTE Distance K,” “intern unnamed min-heap,” and “verbal Fluency” in twenty seconds.
Priority 4 — projects and technology choices Resume-derived
Rehearse one 60-second project door from
IQVIA Deep Research,
IQVIA Hybrid RAG,
Ylogx,
Horizon, or
Argus.
Then choose two why-cards from chapter 14:
LangGraph versus a linear chain,
Azure Hybrid plus Semantic,
Postgres RLS,
Redis cache-aside,
costmap fusion, or
Argus’s 73% gate.
- Answer pain → alternatives → mechanism → failed alternative → honest 10× → locked metric.
- IQVIA uses Azure AI Search plus GraphDB and LangSmith; do not substitute Qdrant.
- Ylogx uses RLS plus three-tier RBAC and Redis; do not invent a cache TTL, p95, shard, or bill.
- Horizon uses ROS2 Python/C++ and native tooling; Java remains the Live Code language.
Exit test:
Explain one discarded alternative and one real failure mode for two different projects.
Priority 5 — Ylogx live path Resume-derived
Speak the complete Ylogx pitch,
then trace browser to data,
RLS versus RBAC,
role-safe Redis, and
one live-issue class.
- Generated SQL runs as the mapped caller role, never table owner or superuser.
- A cache hit for the wrong organization is still a leak.
- Known KPI dashboards stay deterministic; no model call belongs on the sub-210 ms hot path.
- At 10×, expand policy fixtures, pool and cache tests, query work, then ECS desired tasks; do not claim sharding.
Exit test:
Draw CloudFront → ALB → ECS → NestJS/FastAPI → Redis/Postgres and identify where identity, action, row scope, and caching are enforced.
Priority 6 — Horizon, hackathons, and IEDC Resume-derived
Use the Horizon 60-second answer,
then the stale-occupancy failure and
software-first hardware trade-off.
For leadership breadth, use
the eight-hackathon boundary and
IEDC team-member boundary.
- Lock the numbers: 60 FPS, 2M+ points/s, +40% obstacle detection, −55% collision risk, 17th/80+.
- Use “I” for mechanisms personally handled and “we” for the team placement.
- Do not invent a missed event, radio model, sample size, headcount, or mentee count.
Exit test:
Explain how a live-looking video can coexist with a stale costmap and unsafe planner decision.
Priority 7 — PDF-open resume defense Resume-derived
Use chapter 17’s twenty-second bullets,
imprecise-claim defenses,
listed-versus-evidenced skills, and
GitHub mismatch redirects.
- Mechanism first, exact resume number second, evidence boundary third.
- GiftedBooks’s README currently says AegisAI; the Aug 2026 resume wins.
- InstaRecon gets one ethics sentence: security-awareness demo, consent, no production attacks; then redirect.
- Do not turn a resume percentage into a sample size, causal study, percentile, or monitoring vendor.
Exit test:
Answer five random resume bullets in twenty seconds each without adding an unsupported noun or number.
Priority 8 — Java and fundamentals Standard CS
Rehearse chapter 10’s
locals versus fields,
static behavior,
HashMap internals,
equals/hashCode contract, and
PriorityQueue safety.
If there is time, choose one lane:
RAG fundamentals,
networking,
databases,
CNN and Argus, or
OS and Java memory.
Exit test:
Explain one Java trap, one backend mechanism, and one database invariant in under three minutes.
Priority 9 — one class design and one DS-design IE-asked
Use the chapter 06 invariant-first method.
Pick either Amazon Locker or
Parking Lot for class design,
then use Searchable Collection for HashMap, Trie, and optional heap.
Full spoken + DS-table answers for the later pasted LLD/HLD list are
chapter 22 and
chapter 23 — not a Job 10454435 forecast; HLD is not UTA-default.
- Clarify actors, operations, scale, invalid states, and ownership before drawing classes.
- Compile one happy path and one failure path before extracting strategies or interfaces.
- Prepare both full OOD and DS-design; evidence contains both and does not identify “the AUTA R3.”
Exit test:
Draw no more than six types and name the single owner of each mutable invariant.
Priority 10 — remap new evidence before studying it IE-asked
Use chapter 18’s loop map,
true-third table,
OA-as-R1 trap, and
not-my-format filters.
- First ask whether OA was numbered as Round 1.
- Then count live interviews, identify role/geography/YOE, and preserve unnamed wording.
- Only after remapping should a prompt enter the DSA, LLD, Fluency, or STAR lane.
Exit test:
Given any new paste, state “their heading,” “their live count,” and “Adarsh analogue” before discussing content.
4. Morning-of sequence — 45 to 60 minutes
00–05 Say the loop lock: OA not round; R1/R2 done; no 10454435 forecast.
05–15 Speak one STAR; answer why, failure, metric, ownership, and 10×.
15–30 Write one Java hot loop and dry-run it.
30–35 Contrast Distance K with the intern unnamed min-heap.
35–42 Draw Ylogx identity → RLS → role-safe cache path.
42–47 Say Horizon stale occupancy and locked metrics.
47–52 Answer two resume-metric attacks with evidence boundaries.
52–56 Draw one LLD owner and failure path.
56–60 Check audio, camera, editor, Java shell, water; stop studying.
Morning rule:
No new problem, story, framework, or research paste enters the plan.
5. In-room pivot map
Interviewer action
├─ pastes code
│ ├─ tree / Distance K → chapter 05 Distance K; Java first
│ ├─ heap / k-th → chapter 05 dynamic k-th; clarify k updates
│ ├─ grid / reward → chapter 05 grid DP; define count vs max
│ └─ unnamed → name pattern only; never guess an ID
├─ says “GenAI Fluency”
│ ├─ code is also pasted → code first; Fluency sentence only afterward
│ └─ verbal question → IQVIA/Ylogx + verify + should-not boundary
├─ asks behavioral
│ ├─ failure / feedback → Argus 73% quality gate
│ ├─ deadline / pressure → Horizon sequencing; no invented miss
│ ├─ conflict → hybrid vs vector or RLS vs app-only
│ └─ unnamed LP → Ylogx Dive Deep, then Horizon Deliver
├─ asks project depth
│ ├─ current role / POCs → IQVIA
│ ├─ auth / SQL / latency → Ylogx
│ ├─ real time / hardware → Horizon
│ └─ CV quality → Argus
├─ asks design
│ ├─ entities / actors → Locker or Parking Lot
│ └─ add/search/top-K → HashMap + Trie + heap
├─ opens resume or GitHub → chapter 17 evidence boundary and redirect
└─ cites another “Round 3” → chapter 18 remap before accepting relevance
Direct routes:
Distance K ·
heap ·
grid ·
Fluency classifier ·
STAR prompt map ·
project prompt map ·
LLD contents ·
GitHub redirects ·
round remap.
“I will follow the actual prompt rather than the slot label.
I will clarify the contract, state the simplest correct approach, and then optimize.
If this is behavioral, I will use a genuine resume event and keep the metric boundary explicit.”
If interrupted:
stop immediately, answer the new question, and do not finish the memorized paragraph.