Keep numbers and facts straight

GitHub mismatches

If they open GitHub, label the mismatch in the first sentence, then speak the Aug 2026 PDF. Resume wins. There is no merge timeline on the resume to narrate. Disclose AegisAI when they open the GiftedBooks repo or when Earn Trust is the self-criticism beat. Customer Obsession still leads with hours → 3–10 min on the PDF product.

What they might ask

Walk me through this GitHub README.
Is GiftedBooks the same as AegisAI?
What is this InstaRecon / PhiSiFi repo?

Resume-derived laptop-open hygiene. Not a Job 10454435 stem.

What you say

They opened GiftedBooks. That README currently describes AegisAI. That is a mismatch. The resume product is GiftedBooks: VR learning plus grounded PDF question-answering. Doubts that used to take hours became three to ten minutes. Uptime on that product is ninety-nine point five percent. I would fix the README. I am walking the PDF product, not AegisAI features.

They opened InstaRecon or PhiSiFi. That repo is a security-awareness demo. I only work with consent. I do not run production attacks. The work I can defend on this resume is StratifyLabs, Argus, Ylogx, and IQVIA.

They might ask next

Repo / artifact	What it currently is	What you speak
`giftedbooks-ai` README	Describes AegisAI	GiftedBooks VR + PDF RAG from the PDF. Would fix the README. Not the same product in this room.
StratifyLabs GitHub	Default Next.js README	stratifylabs.design / resume: browser inference, −30% ML iteration, marketplace 50+, URDF/WebGL, Gemini in-scene.
Argus / `argus-stream-api-server`	README 404	Resume detector metrics (YOLOv9 73→89% mAP, 24 FPS, then site outcomes after the gate). No routes written in from a missing README.
InstaRecon / PhiSiFi	Off-resume security demo	Ethics one-liner, then redirect. Not an LP. No exploit steps.
Conv-BI / Warpflow	Ylogx-adjacent	Not extra internships. Live intern story remains report builder + SQL RAG + RLS + Redis + dashboards.
`Gstreamer-UDP`	Webcam-stream analogue	Supporting Horizon artifact, not a second product. Not Argus YOLO.
valAgent	Separate ETL-validator	Not IQVIA Deep Research. Does not inherit 200+ websites.
karyanode / Horizon-Website-old / Team-CUSAT / visionlabServer	GitHub-only names	They do not inherit resume metrics.

If they ask	You answer
They stay on GiftedBooks after the mismatch	Earn Trust is one sentence on the README. Customer Obsession stays hours → 3–10 min on the PDF product. No AegisAI feature list.
They open StratifyLabs	The default Next.js README is not the product. Speak stratifylabs.design: −30% iteration, marketplace 50+.
They open Argus and the README 404s	YOLOv9 73→89% mAP, 24 FPS, then site outcomes after the gate.
They stay on InstaRecon	Repeat consent. Redirect to StratifyLabs, Argus, Ylogx, or IQVIA. Stop. Not an LP.

Keep straight

GiftedBooks on the resume is the VR + PDF RAG product. AegisAI is what the current README describes. They are not the same product in this room.
StratifyLabs product is stratifylabs.design and the resume. The GitHub README is a Next.js scaffold.
valAgent does not inherit IQVIA’s 200+ websites.
Haar, YOLOv8, and ByteTrack belong on other repos / Standard CS alternatives. Argus on the resume is YOLOv9.

Tighten

One sentence on the mismatch, then the PDF product, then stop. Earn Trust owns the README honesty. Customer Obsession stays hours → 3–10 min. If they stay on InstaRecon, redirect after the ethics line. Login Tracker is unverified — leave it unless they name an API in Live Code.

Practice STAR plots that are not yours

DO NOT SPEAK AS ADARSH. The older study book’s chapter 13 is a recognition drill, not autobiography. LC 6653463 UTA: fake LP answers die on cross-questions. Weak vs strong shape lives only under this heading. Chapters 3, 4, and 5 do not import these plots into genuine .say.

What they might ask

Tell me about a time you failed / missed a deadline / disagreed with your manager.
Tell me about a time you saved cost, mentored a pod, or missed a competition.

IE-asked prompt families from other loops (LC 8362604 “missed a commitment,” IE.in 2024-grad). Resume-derived genuine redirects. Job 3100855 / Reddit 1s9z862 is a different Job ID. Exponent / IGotAnOffer commercial samples are Class C shapes — prompt family only.

What you say

That plot is not on my resume, so I will not present it as mine. The example I can defend is Argus at seventy-three percent mAP: that number was not a detector I would page a supervisor on. I gated on eighty-nine percent mAP and twenty-four frames per second on fifteen thousand-plus images, then scaled to twenty-plus cameras.

They might ask next

Practice plot — DO NOT SPEAK AS ADARSH	Where the genuine fact lives
Unnamed company. Vague “we shipped.” Lesson is “I learned communication.”	Named intern product on the PDF. Intern title, then the module you owned. One LP. One metric family.
Missed 99.95% SLA, region blip, postmortem, “we missed the month.”	Failure is Argus 73% mAP as an unshippable alerting gate. Ylogx 99.9% uptime held. IQVIA has no uptime percent.
Manager fight, raised voice, skip-level, “the manager respected me.”	Technical disagreement: IQVIA hybrid + semantic + GraphDB vs vector-only; backup Ylogx RLS-in-database vs app-only WHERE. Commit on evals. No named interpersonal blow-up on the resume.
Mentee of twelve, ratings, promotions, retention +20%.	IEDC Tech Team member. Share setup until someone else can `launch`. No mentee count on the resume.
AWS bill saved $40k, finance thank-you.	Frugality is Ylogx Redis −35% bot-DB latency with tier in the cache key. No invoice on the PDF.
Missed ERC, hero weekend, then a miracle finish.	Horizon 17th / 80+ is a team score and it held. Deadline speech is sequencing + early specific risk on the feed and costmap.
Haar cascades → YOLO miracle overnight.	Haar is not on the resume. Argus detector is YOLOv9 73→89% mAP. Haar/HOG are Standard CS alternatives, not a shipped first attempt.
SEO 403 / Search Console / noindex as Dive Deep.	Dive Deep lead is Ylogx Redis + RLS together. SEO 403 is prep-only / chapter 13.
Qdrant gold set of 10k, intern drama, nDCG.	Qdrant is not on the resume. IQVIA retrieval is Azure AI Search Hybrid + Semantic + GraphDB. LangSmith traces. No gold-set size on the PDF.

Keep straight

Missed SLA belongs nowhere on this resume. 99.9% is Ylogx and it held. Failure for alerting is Argus 73% mAP.
Manager fight, mentee of 12, AWS $40k, missed ERC, Haar→YOLO, SEO 403 as Dive Deep, and Qdrant gold set are practice plots. They are not Adarsh’s stories.
If a prompt sounds like those plots: one sentence that the event is not on the resume, then the locked project for that LP from chapter 3.

Tighten

Closing line of the drill, not a new STAR: that event is not on my resume. Then one genuine card. If you catch yourself naming a company that is not on the PDF, stop. There is no log, quote, invoice, or eval set for the left column.

Where each fact lives

If they ask how you measured and the protocol is not on the PDF: the resume number is X; the mechanism was Y; I do not remember the exact protocol. That is the bar. Skills-line tools are inventory. Production internships are the hops you can walk.

What they might ask

Did you use Spring / Kafka / EKS / Qdrant / Haar?
What was the p95, TTL, AWS bill, gold-set size?
Was Ylogx on LangGraph? Did GiftedBooks use LangSmith?

Resume-derived lookup. Other loops are not a forecast for Job 10454435.

What you say

I do not have that protocol on the resume. The number I will defend is the resume metric for this product. The mechanism was the hop I owned. I will not add a baseline N that is not written down.

Internship bullets used Python FastAPI, TypeScript NestJS, and ROS2. Live Code is Java. Spring, Kafka, and EKS are not on those bullets, so I am not describing the internships as if they were those services.

They might ask next

Item	Where it lives
Spring, Kafka, EKS as intern ships	Internship bullets used Python FastAPI, TypeScript NestJS, and ROS2. Live Code is Java. Spring, Kafka, and EKS are not on those bullets. Ylogx serving path is ECS + Docker. Kubernetes is skills-listed. I did not shard.
Ylogx or IQVIA on the JVM / Spring Boot internships	Live Code is Java. Production internships and projects: Python FastAPI / LangGraph / YOLOv9, TypeScript NestJS/React, ROS2. Say the split once, then stop.
LangGraph on Ylogx	Ylogx chatbot is LangChain SQL RAG. LangGraph is IQVIA (Deep Research and adaptive retrieval).
LangSmith on Ylogx or GiftedBooks	LangSmith is IQVIA. GiftedBooks and Ylogx have no LangSmith on the resume.
GiftedBooks = AegisAI	README mismatch. Resume wins. Speak GiftedBooks VR + PDF RAG from the PDF.
Haar cascades as Argus origin	Haar is not on the resume. Argus detector is YOLOv9 73→89% mAP. OpenCV may decode/resize; that is not a cascade jump.
YOLOv8 / ByteTrack / OSNet as Argus	Purplle-style prep-only. Argus on the resume is YOLOv9.
95% mAP	Resume is 73 → 89%. 95 is not on the PDF.
Qdrant	Qdrant is not on the resume. IQVIA retrieval is Azure AI Search Hybrid + Semantic + GraphDB. FAISS is skills-line only.
Missed ERC / missed run / hero weekend	Resume: 17th / 80+. The placement held. Deadline talk is sequencing, not a miss.
Missed 99.9% / missed SLA / region blip	Resume: 99.9% uptime held at Ylogx. Failure for alerting is Argus 73% mAP.
Manager fight / raised voice / skip-level	No named interpersonal blow-up. Conflict is technical: hybrid vs vector-only; backup RLS-in-DB vs app-only WHERE.
Mentee count / pod of 12 / ratings / promotions	IEDC Tech Team member + Horizon sharing. Not a manager. No mentee N on the PDF.
AWS bill / $40k / finance thank-you	No invoice on the PDF. Frugality is Redis −35% bot-DB latency, not dollars.
p95 / cache TTL / Datadog / SLO window / nDCG / gold-set size	Not on the PDF. Mechanism + resume number, then “I do not have the protocol on the resume.”
Auth0 as Ylogx production isolation	Auth0 is skills-listed. Ylogx isolation is RLS in Postgres.
GPU model name / TensorRT / INT8 as shipped	GPU vs CPU is a shape. No SKU on the resume.
Bankers Algorithm as Ylogx	Single GFG last-F2F hit elsewhere. Intern BI is report builder + SQL RAG + RLS + Redis, not Bankers.
Login Tracker (`new_login` / `get_oldest_login`)	Unverified mentor practice. Not a named Job 10454435 question. See unverified.
Rate Limiter as a UTA-default leftover LLD, or as a system you shipped	Independent SDE I live-round count = 1, not UTA two-DSA default. Ylogx Redis is cache. See unverified.
A named Job 10454435 live question	Still none. Unnamed stays unnamed.
What your Round 3 contained	Not written in this booklet. See Round 3.

Listed on skills ≠ shipped on an intern bullet

Redirect to the evidenced hop: Kubernetes, GraphQL, WebSockets, ProtoBuf, Flask, Firebase, Supabase, Prisma, FAISS, Neo4j, TensorFlow-as-the-Argus-jump, GitHub Actions-as-the-product. Prep-only ≠ intern evidence: Purplle YOLOv8+ByteTrack, SEO 403, Django CUSAT portal, abliteration, a Rate Limiter you practiced.

If they ask scale / 10×

Interviewer-quoted “10× the traffic” in opened later-slot bodies this pass: n = 0 (could not verify). LC 8362604 “stories lacked massive scale” is self-critique, not a question. Honest intern line: I did not shard. First: cache / pool / eval / FPS gate / ECS desired count / drop stale frames. 10× users without RLS is 10× leak. 10× cameras at 73% mAP is 10× ignored alarms. 10× BRDs without cite-or-abstain is 10× fluent errors. Would-do stays labeled would-do.

Keep straight

Internship bullets used Python FastAPI, TypeScript NestJS, and ROS2. Live Code is Java. Spring, Kafka, and EKS are not on those bullets.
99.9% uptime is Ylogx. IQVIA has no uptime or latency percent on the resume.
LangSmith and LangGraph are IQVIA. Ylogx chatbot is LangChain SQL RAG.
If the protocol is missing from the PDF, say so. A fake sample size is worse than an honest gap.

Tighten

One hop, one resume number, stop. Would-do is labeled would-do. Rushikesh-style “no massive scale” is not your line. Isolation, a date that does not move, and a quality gate are the challenge.

Metric families

A leftover hour often asks why a metric exists. Mixing two products in one sentence is how the story falls apart. If the percent has no baseline or sample size on the resume: mechanism + resume number + no invented N. Cross-family lock: Horizon 60 FPS is not Argus 24 FPS. Stratify −30% is not FPS. Argus mAP is not −50% violations. IQVIA GraphDB is not LangGraph. Ylogx 99.9% is not GiftedBooks 99.5%. Ylogx sub-210 ms is not GiftedBooks sub-300 ms.

What they might ask

Walk me through that number. How did you measure it?
Is that the same 200+ / the same FPS / the same uptime as the other product?

Resume-derived families. Other candidates reported metric grills (LC 7981646, 8362604). Not a 10454435 forecast.

What you say

Seventy-three percent mAP was not a number I would page a supervisor on. Missed helmets and junk boxes both teach people to ignore the alarm. I gated on eighty-nine percent mAP and twenty-four frames per second on fifteen thousand-plus images, then scaled to twenty-plus cameras. Violations down fifty percent and compliance doubled are site outcomes after that gate, not another name for mAP.

At IQVIA, LangGraph is the agent. GraphDB is the requirement graph I hop for Hybrid RAG. Deep Research’s 200+ is websites ranked. Hybrid RAG’s 200+ is page BRDs. Those are two systems.

They might ask next

IQVIA — two systems, two “200+”

Number	Family	Stays on
200+ websites	Deep Research ranking	Parallel Firecrawl / Bing / DuckDuckGo / Playwright; rank quality, recency, agreement; cite-or-abstain. nDCG and gold-set size are not on the PDF.
200+ page BRDs	Hybrid RAG coverage	Azure Hybrid + Semantic + GraphDB hops + LangGraph adaptive retrieval + LangSmith traces. IQVIA has no latency percent. Redis −35% and sub-210 ms stay on Ylogx.
LangGraph	Agent orchestration	Deep Research and adaptive retrieval. GraphDB is the Hybrid RAG dependency store — not “LangGraph hops.”

IQVIA primaries: Invent and Simplify, Learn and Be Curious, Backbone. IQVIA ranking contradiction is Dive Deep backup, not a fourth Ylogx or IQVIA primary.

Ylogx — keep the poster from eating every LP

Number	Family	Stays on
40% faster reports	Report builder	Custom reports FastAPI / NestJS / Postgres. Not chatbot latency. p95 is not on the PDF.
99.9% uptime	Serving path	ECS desired count, ALB health, CI/CD, CloudFront. GiftedBooks uptime is 99.5%. IQVIA has no uptime percent.
+65% analysis productivity	SQL RAG	NL → SQL as user role, still behind RLS. Not model “accuracy.” Customer Obsession primary stays GiftedBooks; this is the backup.
RLS+RBAC 3 tiers	Isolation	Postgres role, not React hide. Named extra tiers are not on the resume.
Redis −35%	Bot DB latency	Cache-aside, tier in the key, RLS still on. Dashboard path is sub-210 ms. TTL and p95 are not on the PDF. Redis is not the CDN.
30 dashboards +60% ops	KPI vs ad-hoc waits	React/Recharts. Survey N is not on the PDF. 60% is not the chatbot.
sub-210 ms	Dashboard path	Edge + origin + no LLM on that GET. Chatbot is −35% bot DB. GiftedBooks API is sub-300 ms.

Ylogx primaries: Ownership, Frugality, Dive Deep only. Dive Deep Result is Redis and RLS together — not the whole intern poster. SEO 403 is prep-only. CloudFront 403 is not ALB 502, not origin 5xx, not RLS 403.

Horizon vs Argus vs Stratify — FPS and percents

Number	Family	Belongs with
Horizon 60 FPS	GStreamer camera infra	Rover pipeline. Argus live infer is 24 FPS. Stratify −30% is ML iteration time.
Argus 24 FPS	Live industrial infer budget	After 89% mAP. Horizon 60 FPS stays on the rover.
Stratify −30%	ML iteration time (browser vs local GPU loop)	Not a camera rate. Not Argus or Horizon FPS.
Horizon +40% obstacle	Costmap / perception on sensors we had	Not Argus mAP.
Horizon −55% collision	Fusion vs extra hardware	Not Argus −50% violations.
Horizon 17th / 80+	Team ERC score	We. The resume has a placement, not a missed ERC.
Argus 73→89% mAP	Detection quality gate	73% is not shippable for alerting. Not “accuracy,” not 89% of workers, not 95 mAP, not −50%.
Argus 15,000+ images	Training set	Split tables and precision-recall extras are not on the PDF.
Argus 20+ cameras	Scale after 89% + 24 FPS	73% does not multiply to 20+ cameras.
Argus −50% / 2×	Site alerts after the gate	Plant-floor outcomes. Not another name for mAP. Plant headcount is not on the PDF.

Horizon primaries: Bias for Action, Deliver Results, Earth’s Best Employer. Argus primaries: Are Right, Highest Standards, Success and Scale. StratifyLabs primary: Think Big. Marketplace 50+ is models in the catalog, not 50 GPUs per user. Stratify GitHub README is default Next.js — resume / stratifylabs.design wins.

GiftedBooks and education

Number	Family	Stays on
sub-300 ms	GiftedBooks API	Not Ylogx sub-210 ms. IQVIA has no latency percent.
99.5% uptime	This product	Ylogx is 99.9%.
+35% reading / +50% engagement / 2.5× comprehension	Resume claims	Study protocol and student headcount are not on the PDF.
Hours → 3–10 min doubts	Grounded PDF Q&A	Customer Obsession primary. LangSmith is IQVIA. AegisAI features stay off this walkthrough.
CUSAT CSE 8.42 / 10	Education	Do not round up. Not Earth’s Best Employer evidence.
Eight hackathons	CodeRecet 1st, MLH Best Project, Magnathon 2.0 runner-up	IEDC Tech Team — no mentee count. Bias backup, not a fourth Ylogx primary.

Which Leadership Principle uses which product

Bucket	Primaries (exact count)
Ylogx	Ownership, Frugality, Dive Deep (Redis+RLS, not SEO 403)
IQVIA	Invent and Simplify, Learn and Be Curious, Backbone (hybrid vs vector)
Argus	Are Right, Highest Standards, Success and Scale (20+ cameras only after 89% and 24 FPS)
Horizon	Bias for Action, Deliver Results, Earth’s Best Employer
GiftedBooks	Customer Obsession, Earn Trust
StratifyLabs	Think Big
IEDC	Hire and Develop

A backup is a second story, not a fourth trophy on the same intern. Default unnamed pair for writing (lock, not evidence you will be asked): Dive Deep then Deliver Results; third GiftedBooks 3–10 min. If Horizon already appeared in an intro, Deliver backup = Ylogx 40% / 99.9%. Failure / feedback: Argus 73%. Deadline: sequence, ERC held, 99.9% held. Conflict: hybrid vs vector.

Keep straight

73% mAP is a fail for alerting. The gate is 89% mAP and 24 FPS, then 20+ cameras. −50% / 2× are after the gate.
Dive Deep lead is Ylogx Redis and three-tier RLS together. Anecdote: chatbot returned SQL and rows. Reality: every NL turn still hit Postgres, and a generated JOIN could cross org if isolation lived in the app. Measure bot database latency. Tier in the Redis key. A cached wrong-org answer is still a leak.
Deadline: ERC date does not move → 60 FPS feed and costmap first; extra hardware is the cut. If a ship-blocker is late, the sentence is “the feed is late,” or “the costmap is stale,” or “RLS is not in the database yet.” 73% safety models do not scale to buy a date.

Tighten

One spoken STAR, one project, one metric family. If they already heard Ylogx Dive Deep, IQVIA ranking contradiction is backup. If they already heard Horizon Deliver, Ylogx 40% / 99.9% is backup. Stop before the whole intern poster lands in one Result.

I vs we

Saiteja’s later reflection (“I did this” rather than we) is hygiene, not a quote Amazon said, and not a Job 10454435 reject cause. Resume verbs Architected / Built / Designed still sit on an intern title: say intern, then the module. Intern ownership stays I. ERC placement stays we.

What they might ask

What did you do versus the team?
Who put RLS in the database?
Did you place 17th at ERC?

Resume-derived ownership. IE-asked “what did YOU do” as a later-slot shape in other write-ups.

What you say

I was an intern. I put three-tier RLS in Postgres on the shared chatbot and report path. I measured bot database latency and put the tenant tier in the Redis key. Bot database latency dropped thirty-five percent.

On Horizon I owned the GStreamer feed, ZED mapping into the costmap, and the software path that had to hold sixty frames per second. Seventeenth of eighty-plus is the team’s ERC result. I did not design the chassis or the electrical stack.

They might ask next

Claim	Pronoun	Note
RLS / Redis / SQL RAG / report builder / dashboards at Ylogx	I shipped	Intern; still I for the module
IQVIA two FastAPI systems	I architected as intern	Not Head of Platform
Argus train / mAP–FPS gate / NMS / event log	I	Site −50% / 2× involve the plant. Plant headcount is not on the PDF.
GStreamer, ZED mapping, costmap software	I owned the path	Chassis / electrical: we / not-claimed
ERC 17th / 80+	We	Team score
Hackathon named prizes	Team or I per event	Headcount extras are not on the PDF. Bias backup, not a fourth Ylogx primary
IEDC / Horizon knowledge sharing	I shared setup	We for team outcomes. No reports. No mentee N

Keep straight

Ylogx isolation and Redis are I. Horizon 17th is we. Argus detector gate is I; site −50% / 2× are plant outcomes after alerts existed.
IEDC is member, not manager. Earth’s Best Employer is Horizon sharing, then IEDC pairing. CGPA 8.42 is education, not Earth’s Best.

Tighten

One module per sentence. If they ask what the team did: ERC 17th and named prizes are we. Isolation, gates, and camera software stay I. Then stop.

Do not announce Bar Raiser

A leftover hour after two lives may be HM, BR, GenAI Fluency, DSA+LP, or unlabeled behavioral. You treat the hour in front of you. You do not classify the interviewer out loud. You do not announce that you are raising the bar.

What they might ask

How are you treating this hour?
Nothing — they just start. You may hear that they interview across orgs, or that they are not on the hiring team. You may hear nothing.

IE-asked slot shapes from other loops, not a 10454435 forecast. LC 7850431 R4 heading is Behavioural — dump 13 does not upgrade that heading to BR. LC 7563011: last live was HM, no BR named. Taanya: BR then later HM. Third live after two lives is not always BR.

What you say

I am treating this as a full interview. I can walk a resume decision, and I can write Java if you open the editor.

They might ask next

If they ask	You answer
They hint they interview across orgs	Do the hour anyway. No classification out loud.
They name a Leadership Principle	Tell that story. One project.
They paste code	Write Java. Title on the calendar is not the algorithm.
They ask about a previous live	This booklet has no Round 3 transcript. Speak only what you remember from that room. See Round 3.

Keep straight

LC 7850431 Round 4 heading stays Behavioural. LC 7563011 last live stays HM. Those are other loops.
Nisarg all-16 is not a requirement (claimed 1/11 locked hours, plus a calendar clash). Few-deep is the common write-up.
This leftover hour’s type is not a prediction on this page.

Tighten

Do the hour. If they offer remaining minutes for code, take it. If they offer remaining minutes for another LP, take the next locked story. After the call, leave the slot type unannounced.

Recruiter said LP-only — still code

Calendar and mail can be wrong. LP is not a formality. A Fluency title is not permission to skip the editor. Keep one Java DSA pattern warm. Other loops, not a 10454435 forecast.

What they might ask

They paste a tree after an LP-only invite.
The slot is labeled GenAI Fluency and the editor still opens.
Mail had no Live Code link, and they still ask you to code.

IE-asked in the loops below. Reddit 1qj304z comments this miner pass: I could not verify.

What you say

I will write Java. Let me restate constraints and complexity before I type.

Editor has a tree. This is a tree walk at distance K, not a retrieval problem. I will write Java. If we still have time for Fluency after the code, I can walk Hybrid RAG versus a linear chain.

They might ask next

Loop	Framing	Room
LC 7406809 AUTA	HR: only three LP	DSA tree path A/B/C (then LP “formality”). Selected. Self: third “did not look good,” 30 min / multiple approaches.
LC 6806195 AUTA	Mail no Live Code link	BR still mixed LP (feedback / critical issues / beyond task) + Count Number of Nice Subarrays. Selected.
LC 7981646	Told LLD / expected non-tech	Trees+DP; intense later tech. Selected.
Reddit 1qj304z AUTA Chennai	Recruiter LP-only and it was LP	Also exists. Not permission to unprep Java. Comments this miner pass: I could not verify.
LC 7623949	Labeled Gen AI Fluency	Asked Distance K, O(n), no parent mapping — actually DSA. Intern Fluency LC 7643209 min-heap is not Distance K and not this leftover hour.
LC 6461439	Interview 3 LP-only (cleared)	Interview 4 still extra DSA (Distance K + LCA). Company feedback “improve problem-solving” despite the author saying optimal.

GFG 2025 LP-poor was their second live / our R2 analogue, recruiter “unsure due to LP” — not your R4, still a warning that LP is not skippable. Arijit unexpected SD (Tech 2 / our R2 analogue) and igreaper Kafka/B+ (their R3 / our R2 analogue) stay off this leftover CS hour. Unnamed BR families stay unnamed: no stamp of LC 962 / 210 / 62 / 2272 / 128.

Keep straight

If they paste a tree, write the Java tree first. One Fluency sentence only if they still ask after the code.
LC 7406809 Distance K on their BR after an LP-only mail is their third, not a 10454435 list.
OA is not a round. OA-as-R1 remaps their “R4” to our leftover analogue; their “R3” coding is often already-done R2. LC 6800853 grid paths = our R2 analogue.

Tighten

Talk while you code. Full pattern cards are in 6. Coding. If they never ask Fluency after the tree compiles, leave it.

Questions to ask them

End of this leftover hour — not “end of R3.” Other candidates asked about the team, not the loop. Compensation stays off this stage. “What does Amazon do” stays off. Whether this is the Bar Raiser stays off. How you did and when the offer arrives stay off.

What they might ask

Do you have any questions for me?

IE-asked end-of-hour shape. Reddit 1qj304z: after two LP sub-questions the interviewer asked if the candidate had questions; a Prime Video product pitch ended the hour ~25 minutes. That reverse-Q is a warning, not a template. IE.in 2024-grad interviewer “giving you back your 20 minutes” after enough notes.

What you say

What does a typical week look like on this team for a new SDE I?

What separates someone doing well at six months from someone who is not?

What have you seen new grads struggle with on Live Code versus in production?

What has kept you at Amazon?

They might ask next

If they ask	You answer
They offer time for a technical follow-up	You may ask how this team reviews Java versus the intern stacks you actually shipped — once, without a language apology. You may ask what on-call or code-review looks like for a new grad. Then stop.
They answer with a team problem	One follow-up on on-call or SDE I code review. Then stop.
They say they cannot share	Thank them and stop.
Time remains and they open an editor	Switch to Java. Shiwangi’s leftover DSA existed because time remained.

Keep straight

Pick one or two questions, then stop. A product pitch (Prime Video, Alexa, a startup idea) is not a question.
Compensation, competing offers, “is this BR,” “did I raise the bar,” and a disguised status check on the loop stay off this hour.

Tighten

Stop before a second round of questions. If they end early, thank them. Early end is not a script that you failed or passed. Leave the editor compiling if you coded. After the call, wait for the recruiter.

Login Tracker unverified · Rate Limiter count = 1

Mentor practice is not this Job ID. An independent SDE I design hit is not the UTA two-DSA default. If they start a design you have not shipped, label it Live Code and design it there.

What they might ask

Design a rate limiter / oldest-login store / insert plus get-oldest.
Did you ship a limiter at Ylogx?

Standard CS practice. Rate Limiter independent SDE I live-round count = 1 (IE.in 2025-grad, their R2 of an OA+3 system-design loop). LC 8362604 “rate limiting / geofencing” = story follow-ups on a challenged STAR, not a second LLD. Bhavya’s Spring login Controller / Facade / Service / Repository was an HM design in another loop — not this tracker, not R2 LRU.

What you say

I have not shipped that as an intern system. If you want it designed here, I will design it as a Live Code problem and label what is not on my resume.

They might ask next

If they ask	You answer
Insert plus oldest-login	Ask them to name the API. Design it as Live Code. Login Tracker (`new_login` / `get_oldest_login`) is unverified mentor practice. It is not a named Job 10454435 question.
Rate limiter as intern evidence	Ylogx Redis is cache-aside on the bot path, −35% bot-DB latency, tier in the key. That is not a limiter you shipped.
Is this the UTA leftover LLD?	Independent SDE I live count for Rate Limiter = 1, on an OA+3 system-design loop. Not UTA two-DSA default, not a predicted R4 LLD for this Job ID.
Bankers Algorithm on intern BI	Single GFG last-F2F hit elsewhere. Ylogx is reports, SQL RAG, RLS, Redis, dashboards.

Keep straight

Login Tracker: unverified. Volunteer it only if they describe the API in this room.
Rate Limiter: count = 1, not UTA. HLD Kafka novels are not the SDE I UTA default.
Official JD for 10454435 was 404 this research pass. Reddit 1s9z862 = Job 3100855, a different Job ID.

Tighten

If they want it designed, design it and label the gap. Then stop. Pattern cards if the editor opens: 6. Coding.

What Round 3 contained

Round 3 already happened. This booklet has no transcript of that hour and no substitute list. Job 10454435 still has no public named live-round question for any live. Other candidates’ “Round 3” write-ups are their loops. OA-as-R1 posts often number OA as Round 1, so their Round 3 is frequently this loop’s R2 analogue — already done for you.

What they might ask

What did they ask you last time?
How did the previous round go?
Nothing — they start this hour from a blank page.

Resume-derived lock. This page does not fill a gap with someone else’s stem.

What you say

I can talk about this hour. I will walk the prompt in front of me. I am not reconstructing a previous live from a write-up.

They might ask next

If they ask	You answer
What did Round 3 ask?	Speak only what you remember from that room. This booklet has no Round 3 question list. Other IEs’ Round 3 coding is often already-done R2 after an OA-as-R1 remap.
Questions for Job 10454435?	Still none in public write-ups. Unnamed stays unnamed.
They start a new LP or paste code	Answer this hour. Chapter 3 for stories, chapter 6 if the editor opens.

Keep straight

R1: 13 Aug 2026. R2: 18 Aug 2026. R3: done. This folder is Round 4 only.
OA is not a round. LC 6800853 grid paths map to our R2 analogue. Arijit unnamed SD maps to R2. Igreaper Kafka/B+ maps to their R3 / our R2 analogue.
Naina R2 LP felt solid — a falsified “went poorly” stamp stays off. Prem Raj LinkedIn body is Selected. Aditya Devansh reject cause: I could not verify. Dhruv Singhal communication reject: SDE-3 / 5+ YOE / 4-round HLD — not UTA SDE I.

Tighten

This hour is the work. If they want a story, one locked STAR. If they want code, Java. After the call, do not announce to the recruiter what type this leftover hour “was.”

One-line redirects

If a prompt tugs toward a practice plot or a missing artifact, land on the row that is actually on the resume. Then stop.

Temptation	Where it actually lives
Failure / missed SLA	Argus 73% quality gate. Ylogx 99.9% held.
Manager fight	Hybrid vs vector; commit on evals. Backup: RLS-in-DB vs app-only WHERE.
Dive Deep	Ylogx Redis + RLS together. SEO 403 is prep-only.
Deadline miss	Sequence + early specific risk. ERC held. 99.9% held.
Haar / YOLOv8 / ByteTrack	YOLOv9 73→89, 24 FPS. Haar is not on the resume.
Qdrant	Azure AI Search + GraphDB. Qdrant is not on the resume.
AegisAI README	PDF GiftedBooks.
InstaRecon internals	Consent demo, then redirect. Not an LP.
10× shard	Cache / pool / eval / desired count. I did not shard.
Spring / Kafka / EKS internships	Python FastAPI, TypeScript NestJS, ROS2. Live Code is Java. Those three are not on the intern bullets.
LangGraph on Ylogx	Ylogx is LangChain SQL RAG. LangGraph is IQVIA.
LangSmith on Ylogx / GiftedBooks	LangSmith is IQVIA.
LP-only mail	Keep Java warm (7406809, 6806195).
Fluency chat + tree paste	Code Distance K first.
“What did R3 ask?”	This booklet has no transcript. Speak only that room.
“Questions for 10454435?”	Still none.
Login Tracker / Rate Limiter as your R4	Unverified / count = 1, not UTA.

Tighten

Genuine spoken cards: 3. Stories. Mechanisms: 5. Projects. Java if the editor opens: 6. Coding.