8. Leadership Principles — Round 3 and 4

Spoken STAR for Adarsh Vishwakarma. Later lives grill failure, feedback, and deadlines. Scan this chapter, then chapter 12. Full spoken hour: _answers/05-star-br.md. Study book 16-LP: study book chapter 11. Do not break the spread. Job 10454435 still none. Labels: IE-asked Resume-derived.

Situation 15s, Task 10s, Action 40s (three steps with a mechanism), Result 15s then stop. They will ask why three times. Answer the why. Use I. No invented miss, no fight scene, no Haar, no Qdrant, no SEO 403 as the Dive Deep lead.

Primary spread (do not break)

LPPrimaryBackup
Customer ObsessionGiftedBooksYlogx +65%
OwnershipYlogx RLS + 99.9%Horizon software
Invent and SimplifyIQVIA Hybrid RAGStratifyLabs browser
Are Right, A LotArgus 73%→89%Ylogx measured latency
Learn and Be CuriousIQVIA LangGraphHorizon ROS2
Hire and DevelopIEDC Tech TeamHorizon teammates
Highest StandardsArgus mAP gateYlogx 99.9% / 210ms
Think BigStratifyLabs marketplaceIQVIA 200+ sites
Bias for ActionHorizon ERC datehackathons / Redis ship
FrugalityYlogx Redis −35%Horizon fusion vs hardware
Earn TrustGiftedBooks RAG + 99.5%Ylogx 3-tier RLS
Dive DeepYlogx Redis + RLSIQVIA retrieval
BackboneIQVIA hybrid vs vector-onlyYlogx RLS-in-DB
Deliver ResultsHorizon 17th/80+Ylogx 40% / 99.9%
Earth’s Best EmployerHorizon knowledge sharingIEDC
Success and ScaleArgus 20+ cameras after 89%Ylogx ALB/ECS with RLS

Ylogx already has three primaries (Ownership, Frugality, Dive Deep). IQVIA three (Invent, Learn, Backbone). Argus three (Are Right, Highest Standards, Success and Scale). Horizon three (Bias, Deliver, Earth’s Best). GiftedBooks two (Customer, Earn Trust). Do not promote a backup into a fourth primary.

Default unnamed pair

If they do not name the LP, lead Dive Deep then Deliver Results. Third: GiftedBooks 3–10 minutes.

Q. Tell me about a time you had to dive deep IE-asked

Situation. At Ylogx the SQL RAG chatbot returned rows on a demo. Two problems hid under that. Every natural-language turn hit Postgres, so latency felt like the model thinking. And if isolation lived only in the application, a JOIN could return another organization’s rows.

Task. Find whether slowness was query shape, missing cache, or security in the wrong layer, and make three organizational tiers real on the path the bot and the reports already shared.

Action. I measured bot database latency instead of blaming the generator. I put RLS and RBAC in Postgres so generated SQL ran as the user’s database role. I put Redis on the hot path for a 35 percent drop without skipping RLS. A cached answer still had to be the right tier.

Result. Minus 35 percent bot database latency, plus 65 percent analysis productivity, 40 percent faster reports, 99.9 percent uptime, sub-210 milliseconds on dashboards. I will not lead this with SEO 403.

Why Redis not a bigger RDS? Repeat NL questions are cache-shaped. Why still RLS after cache? A cached answer for the wrong org is a leak.

Q. Delivered results under a hard constraint IE-asked

From February to June 2024 I was Horizon core software for ERC 2024. More than eighty teams. The date does not move. I owned the path that had to run: GStreamer at 60 FPS, ZED 2 mapping at more than two million points per second, costmap planning and fusion on the sensors we had. Stale occupancy was the ship-blocker. We placed 17th of more than eighty. Obstacle detection up 40 percent. Collision risk down 55 percent. I will not invent a missed ERC.

If they already heard the rover in the intro, switch to Ylogx 40 percent / 99.9 percent.

R3 / HM / BR prompts from the bible

Q. Last negative feedback / mistake IE-asked

GFG BR; Rudraksh BR. Honest on-resume version is Argus.

73 percent mAP was the feedback. At that number you miss helmets and you fire junk boxes. Either one trains people to ignore PPE alarms. I did not treat 73 percent as a blog metric. I trained on more than 15,000 images, iterated YOLOv9 to 89 percent mAP, required 24 FPS, then scaled to more than twenty cameras with Postgres logs. Violations down 50 percent, compliance doubled, after the gate. I do not have a manager quote. I will not invent one.

Q. Conflict with a coworker or manager IE-asked

Technical. No invented fight. IQVIA: vector-only vs hybrid+GraphDB. Commit on evals. Backup: RLS-in-DB vs app-only.

I do not have a named interpersonal blow-up on the resume. The conflict I will speak is retrieval. The fast path for RAG on PDFs is a vector index. Two-hundred-page BRDs have clause IDs cosine misses. I stated that similar embeddings are not coverage. I argued for hybrid plus semantic Azure AI Search plus GraphDB plus traces. Once we chose the path, I committed by instrumenting evals and test-case generation instead of reopening the tool debate every week. Disagree on retrieval. Commit on evals.

Q. Missing a deadline IE-asked

LC 8362604 BR; several loops. Resume does not state a miss.

I will not invent a missed ERC or a missed 99.9 percent. Amazon wants how you flag risk. ERC was fixed. I sequenced 60 FPS and the costmap so the run happened. Communication if a ship-blocker is late is early and specific: the feed is late, or the costmap is stale, or RLS is not in the database yet. Then we cut extra hardware. We do not cut isolation. We do not scale a 73 percent safety model.

Q. Complex problem, POCs, multiple solutions IE-asked

LC 7563011 AUTA HM.

IQVIA Deep Research and Hybrid RAG. One scraper is not enough — Firecrawl, Bing, DuckDuckGo, Playwright fail on different sites. Vector-only is not enough on two-hundred-page BRDs. A linear chain cannot retry or hop a requirement graph. Stuffing the PDF dies on tokens. LangSmith was the stop: if a generated test case has no overlapping span, the path is not done. I discarded Qdrant as a story because the resume is Azure AI Search. Result: more than two hundred sites ranked, more than two-hundred-page BRDs with traces. No extra percentage.

Q. Significant technical challenge / limited information IE-asked

GFG April 2026 R3/HM. IQVIA Hybrid RAG / ranking contradiction. Two high-ranked pages disagree; first-hit retrieval is false certainty. Agreement plus cite-or-abstain.

Q. Strict deadline / bias for action / outside scope / issue not your task IE-asked

Horizon for date and action. Ylogx RLS for “not the chatbot demo ticket.” Same stories as chapter 11. Do not reuse Ylogx three times in Aditya’s trio: Ownership Ylogx, pressure Horizon, learn IQVIA.

Q. Proud of / current role / learned something new IE-asked

Proud: pick one — Horizon 17th or GiftedBooks 3–10 minutes. Current role: IQVIA two systems. Learn: LangGraph / Azure hybrid, traces as stop condition.

Sixteen LPs — 20-second primary hook

  1. Customer Obsession — GiftedBooks doubts hours → 3–10 min, RAG on their PDF, PYQ topics.
  2. Ownership — Ylogx three-tier RLS in the database, intern, 99.9% held.
  3. Invent — IQVIA hybrid+graph+test-case gen, not a prompt pile.
  4. Are Right — Argus 73% was a fail; 89% + 24 FPS before twenty cameras.
  5. Learn — LangGraph state, hybrid retrieval, LangSmith pass/fail.
  6. Hire and Develop — IEDC / Horizon sharing launch order. Not a manager. No mentee count.
  7. Highest Standards — same Argus gate.
  8. Think Big — StratifyLabs marketplace + browser inference, −30% iteration.
  9. Bias for Action — ship 60 FPS + costmap, do not wait on extra sensors.
  10. Frugality — Redis −35%, not a bigger RDS first.
  11. Earn Trust — GiftedBooks grounded Q&A; Ylogx isolation.
  12. Dive Deep — latency plus RLS, not SEO.
  13. Backbone — hybrid vs vector-only, then commit on evals.
  14. Deliver Results — Horizon 17th/80+.
  15. Earth’s Best Employer — pair on launch files so knowledge is not in one head.
  16. Success and Scale — do not multiply a 73% model across twenty cameras.

Room checklist