This is a rehearsal notebook for Adarsh Vishwakarma. Read it as yourself. R1 was 13 Aug 2026. R2 was 18 Aug 2026. R3 is done. This folder is the leftover hour only. Live Code is Java. Job 10454435 still has no public write-up that names a live-round question. Other people’s leftover hours show what a later slot can look like. They are not a list for this Job ID. Unnamed stays unnamed. This chapter does not describe Round 3.
The calendar may say Bar Raiser, Hiring Manager, GenAI Fluency, Behavioural, or nothing. Other SDE I write-ups used all of those labels after two lives. The room still starts with one prompt. You answer that prompt. You do not name the slot out loud.
What they might ask
Tell me about yourself / walk me through your resume.
A named Leadership Principle, or a work deep-dive.
A pasted problem in Live Code.
Nothing that matches the invite label.
IE-asked as loop shape, not as this Job ID. LC 7563011 AUTA Bengaluru: third live was Hiring Manager, no Bar Raiser named, then an offer call. LC 7850431: Round 4 heading is Behavioural — that heading stays Behavioural.
What you say
I’m here for a sixty-minute interview. I’ll take the first prompt as given and go deep on it.
They might ask next
If they ask
You answer
A named Leadership Principle
The primary story for that principle (chapter 3). One project.
They paste code
Write Java. Restate constraints first.
Current role
IQVIA, Kochi, two FastAPI systems, then stop. Chapter 2 has the spoken paragraph.
What this round “is”
You do not classify it. You answer the prompt they just gave.
Keep straight
Job 10454435 still has no public named live-round question. Other loops are shape evidence.
A leftover hour is not always Bar Raiser, always GenAI, or always more DSA. Those are three shapes other people reported, not a pick for this Job ID.
The calendar label stays unspoken. You do not open with “this is the Bar Raiser.”
This booklet does not describe what Round 3 contained. Round 3 already happened.
Tighten
Stop after the first prompt’s answer. The hour script is chapter 2. Stories are chapter 3. Coding families are chapter 6.
Shapes other candidates reported
A leftover hour after two lives is not one format. Other candidates reported LP-only, LP plus DSA, work deep-dive then leftover DSA, Hiring Manager with no DSA named, and Hiring Manager that was coding. Length was often the calendar hour. Several write-ups ended in about 22–40 minutes. n below is a count in an 11-IE hour lock, not a rate, and not this Job ID.
What they might ask
LP-only for most of the hour.
DSA first, then a couple of Leadership Principles.
About thirty minutes of stories and thirty minutes of code.
Work and stack for ~40 minutes, then a leftover problem because time remained.
IE-asked hour shape. Locked later-slot dump: 11 must-open IEs, not a census. Zero public hours for Job 10454435. Nisarg exact stems stay unnamed.
What you say
I prepared to go deep on one story and to write Java if an editor opens. I’ll follow whichever they start.
They might ask next
If they ask
You answer
The hour stays on stories
Mechanisms and resume numbers. Same story on the why.
DSA after twenty minutes of work talk
Switch to the editor without a second intro. Java.
They end at minute twenty-five
Ask the team-week questions (chapter 2), thank them, stop.
Write-up fact. Not a rule that yours will end early.
Bookstore word-count on Shiwangi’s loop is LLD in an earlier live, not leftover DSA. Independent Rate Limiter SDE I live-round count stays 1, not UTA-default leftover LLD.
Keep straight
n=11 is a lock, not odds for Job 10454435.
Five of eleven hours in that lock were LP-only. Java still belongs in Live Code if an editor opens.
Combine Garlands on 8362604 is the Hiring Manager hour, not the Bar Raiser hour.
Maximum Sum Linked List on 6881058 stays unnamed as an LC numeric id. 962 and 210 are not that leftover title.
Tighten
Use this table to stay ready for more than one shape. Do not pick a winner for tomorrow. If they start a story, stay on the story. If they paste code, chapter 6.
How other loops branched after two lives
You cannot know this leftover hour’s label from other people’s posts. You can know what those posts did after two lives, then stay ready for more than one shape. The tree is a reading aid. It is not a forecast for Job 10454435.
What they might ask
A slot labeled Bar Raiser that is the last named live, or a Bar Raiser then a later Hiring Manager.
A slot labeled Hiring Manager that is the last named live, or an HM then a later Bar Raiser.
A heading that is only Behavioural, with no Bar Raiser upgrade in the post.
IE-asked as loop shape. HM-vs-BR dump: nine unique opened bodies, not a census. AUTA first-hand in that table: five reports, not a process spec.
What you say
I’ll treat whatever they open — a story, a tree, or a work walkthrough — as the round.
They might ask next
If they ask
You answer
HM-shaped: POCs, multiple solutions, current work
IQVIA Hybrid RAG plus LangSmith traces (Invent / Backbone). Chapter 3 and 5.
BR-shaped LP, no name
Default pair: Dive Deep (Ylogx Redis + RLS), then Deliver (Horizon 17th). Chapter 2.
A later invite still arrives after this hour
That also happened on other loops (8362604, 7724048, Taanya). Answer the new hour the same way.
Reading tree (other candidates, OA omitted)
After two lives, other candidates’ next named slot:
labeled Bar Raiser
├─ last named live
│ LC 6806195 LP + Nice Subarrays (mail had no Live Code; DSA anyway)
│ LC 7406809 tree path, then couple LP (HR had said three LP only)
│ Reddit 1qj304z LP unnamed, ~25 min (recruiter said LP only; body reports LP)
│ Prince Medium internships + LP unnamed — Job 3057703, not this Job ID
└─ then a later HM
Taanya BR deep-dive, then final HM maps+heaps unnamed + GenAI
labeled Hiring Manager
├─ last named live
│ LC 7563011 POC/research prompt, no DSA named, no BR named, offer after
└─ then a later BR
LC 8362604 HM Anagrams variation + Combine Garlands, then BR LP 1 hour
LC 7724048 HM Next Greater + LLM prompts, then BR three LP scenarios
neither HM nor BR on the heading
LC 7850431 R3 GenAI+tech, R4 Behavioural (heading stays Behavioural)
TUF popular-companies: their numbered Round 4 after OA-as-R1 is usually
the third-live analogue. A true fourth live showed as their Round 5 on
four IEs in that folder.
Keep straight
These n values are not probabilities for Job 10454435.
LC 7563011 named no Bar Raiser. The post does not prove a hidden one.
LC 7850431 Round 4 heading is Behavioural. That heading is not Bar Raiser.
Absence of a named fourth on many AUTA third-was-BR loops is not proof Amazon never schedules one.
Prince is Job 3057703. That requisition is not 10454435.
Tighten
Read the tree once. In the room, follow the first prompt. TUF “Round 4” after OA-as-R1 is usually already-done analogue, not leftover LLD.
The OA-as-R1 remap
This card is for how you read other people’s “Round 3” and “Round 4.” If their Round 1 was the online assessment, every later number is shifted. Their Round 3 is often the second-live analogue — already done in this loop — not the leftover hour. OA is not a round. You do not say that sentence in Zoom.
What they might ask
Nothing in the room. This remap is study-time.
A write-up titled Round 3 that is Kafka, grid paths, or Rotten Oranges after OA numbered as Round 1.
Rule from the later-slot merge. Worked examples: GFG 2025, LC 6800853 / 6881058, igreaper. TUF email keyword “their R4 SD” remaps to a third-live analogue, not a guaranteed leftover LLD.
What you say
I answer the prompt in this room. I do not number other people’s rounds out loud.
They might ask next
If they ask
You answer
(Study-time) their heading is OA as Round 1
Drop OA. Count lives. Decide whether the heading is leftover-shaped.
(Study-time) their Round 3 is Kafka or grid paths after OA-as-R1
Second-live analogue. Already done in this loop. Not tonight’s leftover grind.
(Study-time) TUF live Round 4 after OA-as-R1
Usually third-live analogue. Their Round 5 on four IEs was a true fourth.
Usually the third-live analogue. Their Round 5 on four IEs was a true fourth.
Keep straight
OA is not a round. That sentence is for reading write-ups, not for the interviewer.
Arijit’s unnamed system design (Tech 2) is not leftover-shaped from this remap.
Intern Fluency min-heap (LC 7643209) is intern loop, not this leftover hour.
Tighten
When a new write-up appears tonight: drop OA first, count lives, then decide. The next three cards are worked examples.
GFG 2025 leftover stems
A GFG Off-Campus 2025 write-up numbers OA as Round 1, then two technicals, then a Bar Raiser. The Bar Raiser is leftover-shaped. Their Round 3 coding (Connect Sticks + Rotten Oranges) is second-live analogue — already done in this loop. Recruiter comment that the interviewer was “unsure about my candidacy due to the LP part” was after that second technical, not the Bar Raiser.
What they might ask
Current working role and responsibility.
Last time you had to deep dive into a particular bug or task.
Last time you had a conflict with a co-worker/manager.
How do you handle feedback, and when was the last time you received negative feedback?
The last time you learnt something that wasn't required at your job, what was your way of learning, and how much time did it take?
Source: Amazon Interview Experience for SDE-I (Off-Campus)(2025). OA-as-R1 Y. UTA/AUTA not named. Job ID on the page is not 10454435. IE-asked on the BR list. Follow-up tree beyond those five stems: not in the opened body. Resume-derived answers below.
What you say
I am a software developer intern at IQVIA in Kochi. I work on two systems — LangGraph Deep Research across more than two hundred websites, and Hybrid RAG on Azure AI Search with a GraphDB for two-hundred-page BRDs, with LangSmith tracing. I stop there.
They might ask next
If they ask
You answer
Deep dive into a bug or task
Ylogx Redis + RLS together (Dive Deep primary). Not a second IQVIA intro.
Conflict with a co-worker / manager
IQVIA hybrid versus vector-only, then commit on LangSmith evals (Backbone). No manager quote on the resume.
Negative feedback
Argus 73% mAP was not shippable for floor alerts. Gate was 89% mAP and 24 FPS, then 20+ cameras.
Learnt something not required
IQVIA LangGraph traces as the stop. If Invent, Learn, and Backbone already used IQVIA, this goes to Horizon ROS2 + GStreamer + ZED (Learn backup).
Keep straight
Their Round 3 coding is second-live analogue. Leftover grind is the five BR stems, not Connect Sticks.
Current role is IQVIA, April 2026 to present, two FastAPI systems. Horizon and Ylogx wait until they ask what you did before.
Hours spent learning are not on the resume. Speak the mechanism (traces as stop).
GFG 2025 is not this Job ID.
Tighten
Current-role answer stops after two IQVIA systems. The next stem is a new story, not four internships stacked onto IQVIA.
LC 6800853 / 6881058
Two Discuss posts are the same candidate. The earlier post numbers a DSA hour as a later round. The later post is a Bar Raiser with a linked-list construction. OA-as-R1 applies. After dropping OA, that Bar Raiser is a third live, leftover-shaped — not a stamped problem for this Job ID.
What they might ask
Maximum Sum Linked List from two sorted lists with common nodes (candidate wording; no LC numeric id linked for that BR item).
Constraints: negatives? common = equal value or same node? strictly increasing?
LC 6800853 (grid paths + digit-subtract family) now hrefs LC 6881058. Title AUTA on 6800853. FLAG OA-as-R1. Chronological lives after dropping OA: the BR is the third live. IE-asked as shape. Candidate wording on the BR item stays unnamed as an LC numeric id. 6800853 grid paths = second-live analogue, already done. LC 6369243 is a different BR (third live, OA-as-R1 N) with similar linked-list maximum-sum wording; still no 962 stamp.
What you say
I build one walk that always takes the larger remaining sum, switching lists only at a shared node, and I track the running sum in a long so the total does not wrap.
They might ask next
If they ask
You answer
Can values be negative?
Ask before coding. If yes, “always take larger remaining” still holds; dry-run a negative tail.
Does common mean equal value or same node identity?
Ask. Implement the one they confirm.
Are the lists strictly increasing?
Ask. It changes how you detect the next shared node.
Extra space is banned
Two pointers on the live lists. Not a copied array.
Keep straight
6800853 grid paths belong to the second-live analogue. Leftover shape evidence here is 6881058’s list construction.
The candidate did not link an LC numeric id for the BR item. 962 and 210 are not that title.
This is leftover shape evidence for this booklet. It is not “the” Job 10454435 question.
Tighten
Dry-run a tiny pair of lists out loud, then type Java. Full families live in chapter 6.
igreaper leftover why-chain
Pratham Sharma’s Medium post numbers OA as Round 1. Their Round 3 includes Kafka ordering and B versus B+ trees. That hour is second-live analogue. Their Round 4 Bar Raiser is leftover-shaped LP with deep why-chains. “How did you scale it?” on that IE was their Round 2 (first live after OA), not the Bar Raiser.
What they might ask
Why did you make that decision?
What was the measurable impact?
What would you do differently?
How did you handle disagreements?
How did you measure success?
Source: Amazon SDE-1 Interview Experience (3 Mar 2026). OA-as-R1 Y (Round 1 = OA Oct 2025). Loop: OA · Round 2 Nov · Round 3 Dec · Bar Raiser Round 4 Jan 2026 · offer Feb. UTA/AUTA not named. Not Job 10454435. IE-asked on the BR why-list. Project discussion themes: technical depth, trade-offs, scaling decisions, failure handling, metrics — themes, not a quoted “10×” question. Interviewer-quoted 10× in opened later-slot bodies: n=0.
What you say
I pick one project and stay on it. I can say why I chose the mechanism, what number moved, what I would sequence earlier next time, and where I disagreed technically and then committed.
They might ask next
If they ask
You answer
Why, on the story you just told
Same project. Mechanism, discarded option, number that moved.
What would you do differently
Same story: denied-tier tests earlier (Ylogx), gold questions frozen earlier (IQVIA), ship-blocker named in week one (Horizon).
Disagreements
Hybrid versus vector (IQVIA Backbone) or RLS-in-database versus app-only WHERE (Ylogx backup). Then commit.
Keep straight
Kafka ordering and B versus B+ on this post are their Round 3 — second-live analogue, not leftover CS.
A “10× users” interviewer quote was not found in opened later-slot bodies (n=0). Honest scale talk is Argus 20+ cameras after 89% mAP / 24 FPS, or Ylogx ECS with RLS already on, and “I did not shard.”
This IE’s Round 2 architecture list stays in Round 2 analogue, not in the leftover why-tree.
Tighten
Answer the why on the same story. A new internship because they asked “why” is a restart. Stop after the mechanism plus one number.
Recruiter mail versus the room
The mail or the recruiter call is not the room. Two AUTA Bar Raisers were told LP-only or given no Live Code link and still received DSA. Two other leftover hours were told LP-only and were LP. Java stays warm either way.
What they might ask
Leadership Principles only — then they paste a tree or an array anyway.
The invite has no Live Code link — then they still want code.
The invite said LP-only — and the hour is LP.
IE-asked framing class: n=2 AUTA where “no DSA” failed. Not a rate. LC 7981646 (told LLD / expected non-tech, still trees+DP) is later-slot adjacent; that body was not re-verified this pass — known-bank, not a new stamp.
What you say
If an editor opens, I restate constraints and write Java. If they stay on stories, I stay on stories.
They might ask next
If they ask
You answer
7406809-shaped extra-space ban on a tree
LCA plus DFS on the tree, not a parent HashMap, if they refused extra memory.
6806195-shaped mixed hour
Finish the LP they are in, then switch cleanly to the array. Java.
Couple LP after code, “formality”
A real mechanism — Ylogx Redis + RLS or Horizon 17th — not poster names.
DSA first: tree path source→target printed A/B/C (~30 min to extra-space / LCA). Then a couple LP unnamed “formality.” Selected. Candidate self-note: third did not look good.
Matched. ~22 min LP-only. Candidate notes India BRs sometimes add DSA to fill the hour; this EU hour did not.
Keep straight
Matching LP-only hours (1qj304z, 7623949) are not permission to skip Java rehearsal.
The recruiter being wrong is not a sentence you say in the room.
Live Code language is Java. Internship bullets used Python, TypeScript, and ROS2.
Tighten
Walk in ready for a story and for an editor. If the editor never opens, stay on stories. Chapter 2 repeats the switch.
Hiring Manager versus Bar Raiser
Hiring Manager and Bar Raiser are different labels other people used. Contents overlap: both can be LP, both can be DSA, both can mix GenAI talk with code. You do not need to classify the interviewer. You need the prompt. LC 7563011: HM was last, no BR named. LC 7850431: Round 4 heading is Behavioural.
What they might ask
Can you describe a complex problem you encountered that required in-depth research, development of proof of concepts, and exploration of multiple solutions to address the issue? (LC 7563011; candidate tagged Deep Dive / Earn Trust / Customer Obsession.)
Work deep-dive, then leftover code because time remained.
Story probes such as “was that actually challenging,” rate limiting, geofencing — as follow-ups on a story, not as Rate Limiter LLD.
IE-asked as loop shape. 7563011 interviewer: 15 years (11 at Amazon, SDM). Heading 35 minutes; body ~25 minutes of work. 8362604 BR probes on the first story: whether it was actually challenging; rate limiting decisions; geofencing; trade-offs. Independent Rate Limiter SDE I count stays 1. GFG Apr 2026 HM page heading “Common Questions” — caveat, not “I was asked.”
What you say
Situation. At IQVIA I needed retrieval that still worked when a two-hundred-page BRD and a web ranking disagreed.
Task. Compare approaches and stop on something I could fail with a test, not a fluent paragraph.
Action. I compared one scraper versus parallel tools, a linear chain versus LangGraph, vector-only versus hybrid plus GraphDB, and stuffing a two-hundred-page BRD versus retrieve-and-cite. LangSmith traces were the stop: a fluent answer with no overlapping span failed the test.
Result. Hybrid RAG on Azure AI Search with GraphDB and LangSmith. Resume has coverage (200+ page BRDs), not an IQVIA latency percent.
They might ask next
If they ask
You answer
Was that actually challenging / show the failure
Same IQVIA path: clause IDs, span overlap. Not a new org.
Rate limiting or geofencing on a story
Trade-off on that story if it applies. Redis at Ylogx is cache with the tenant tier in the key, not a Rate Limiter you interned.
Invent already used IQVIA
Simplifying a process → StratifyLabs browser inference −30% and marketplace 50+ (Invent backup).
Geofencing is not on the internship bullets. Stay on the story they challenged.
99.9% and sub-210 ms are Ylogx. IQVIA has no uptime or latency percent on the resume.
GraphDB stores requirement dependencies. LangGraph is workflow state.
Tighten
One complex-problem prompt is one project. Stop after traces-as-stop plus the 200+ page BRD family. Scale, if they ask, is Argus 20+ cameras after the gate, or Ylogx ALB/ECS with RLS already on.
When a later live still arrived
Several AUTA posts label Bar Raiser as Round 3 — the third live. That is hour-shape evidence, not proof this leftover hour is a Bar Raiser. A smaller set of write-ups actually had a fourth live after three already existed. Many AUTA third-was-BR loops named no fourth. LC 7563011 HM was last as a third.
What they might ask
Another story because time remains.
An editor because time remains (Shiwangi: work ~40 min, then unnamed sliding window).
LP names only: Why Amazon; Deep Dive; Learn & Be Curious; Earn Trust (LC 7623949, ~22 min, names not full STAR sentences).
IE-asked as timing/shape. True fourth (or fourth-ish) in the hour lock: LC 8362604 BR after a separate HM — fourth live, LP-only, full hour. Shiwangi Kumari R4 17 Dec 2024 17:00–18:00 — UTA named, not labeled BR/HM. Medium comment “no LP each round”: not re-verified (Cloudflare). LC 7623949 BR after an extra Gen AI Fluency hour. TUF: four IEs showed a true fourth as their Round 5 after OA-as-R1 (max-size subarray k; Longest Consecutive Sequence; HM no coding; behavioral).
What you say
I’ll use the full time they give me. If they say there is time left, I’m ready for another story or for an editor.
They might ask next
If they ask
You answer
Shiwangi-shaped leftover code after stack talk
Restate constraints, then Java. Unnamed sliding window stays unnamed.
7623949-shaped LP names
Dive Deep = Ylogx Redis + RLS. Learn = IQVIA LangGraph. Earn Trust = GiftedBooks RAG + 99.5%. AegisAI README only if they are on GitHub.
TUF Round 5 analogue (true fourth)
Same as any leftover: story or Java. Not a 10454435 list.
Keep straight
“Bar Raiser as Round 3” posts describe a third live. They do not prove this leftover hour’s label.
This booklet does not fill in Adarsh’s third live to complete someone else’s timeline.
GiftedBooks 99.5% is that product. Ylogx 99.9% is Ylogx.
Tighten
If they offer remaining minutes, take the next locked story or the editor. Do not restart the first STAR.
A Fluency label can still be a tree
A slot titled GenAI Fluency is sometimes RAG talk. On at least one opened body it was Distance K with an O(n) ask and no parent mapping. If they paste a tree, the tree is the round until they ask something else. Mixed HM/Fluency hours still pasted code: Next Greater (7724048), LCA + process scheduler (8014509; scheduler wording not re-verified this pass), dynamic k-th largest (7850431).
What they might ask
Nodes at distance K on a binary tree, O(n), no parent field on the node (LC 7623949 Fluency; unnamed LC id).
After the code: how you use models at work.
LC 7623949: labeled Gen AI Fluency, actually DSA. GenAI prompt not named on that Fluency hour. BR after = LP-only ~22 min. AUTA not named (India onsite then EU BR). IE-asked. Siblings: LC 7584467 “GenAI Essential + Coding” — candidate: “just another DSA round.” LC 6461439 Interview 4: nodes at distance K + LCA, unnamed titles, fourth live, rejected. Intern Fluency LC 7643209 min-heap is not Distance K and not this leftover hour.
What you say
This is a tree walk. I’ll build neighbor links from child pointers, then BFS from the target until distance K, and return those nodes. I am not using a parent field on the node.
They might ask next
If they ask
You answer
Time / space
O(n) time, O(n) extra for the adjacency list and the BFS queue.
They still want GenAI after the code
I use models to draft; I verify with traces, tests, or a span that must overlap the source. One sentence, then stop.
They never ask GenAI
Leave it. No RAG lecture.
Keep straight
The Fluency hour on 7623949 was described as similar to nodes at distance K. The candidate did not link an LC id. The stem stays unnamed. 863 is not that title.
Intern min-heap Fluency (7643209) is a different loop.
IQVIA LangGraph is not the answer to a binary-tree prompt.
Tighten
Write the tree first. Chapter 2 repeats the room action. Chapter 7 is GenAI only if they still ask after the code.
Loops that are not this leftover hour
Some posts look like leftover hours and are not this loop. They sit here so they do not leak into tonight’s script. Intern Fluency, SDE II HLD, US/FTC loops, other Job IDs, year-out GFG, and Class C guides are in this bucket.
What they might ask
Nothing from this table as a Job 10454435 leftover default.
If a practice LLD from a guide appears in the editor anyway: Standard CS, solved as Java.
NMF = not this leftover default. Login Tracker (new_login / get_oldest_login) is unverified mentor prep. Rate Limiter independent SDE I live count = 1.
What you say
I answer from internships and projects I actually shipped. I do not borrow another job’s loop as my story.
They might ask next
If they ask
You answer
A practice LLD from a guide appears in the editor
Solve it as Standard CS. Java.
Whether you shipped a Rate Limiter at Ylogx
Redis there is cache with the tenant tier in the key. Independent SDE I Rate Limiter live count stays 1, not UTA-default leftover LLD.
Login Tracker
Unverified. Not volunteered as this Job ID.
Out of scope for this leftover default
Class
Examples
Why not this leftover hour
INTERN
LC 7643209 Fluency min-heap; TUF intern Finals
Intern first/last live. Not SDE I leftover.
SDE II / L5 / SDE-3
SDE II HLD; Dhruv Medium communication reject (5+ YOE, 4-round HLD)
Wrong level. Full retail HLD is not UTA-default SDE I.
US / FTC / Canada AUTA
US 3×60; FTC 5-round; Jay Patel Canada AUTA cart/checkout + LRU-like
Region / contract NMF for this ADCI Karnataka loop.
Other Job IDs
Prince Job 3057703; Reddit 1s9z862 Job 3100855
Different requisitions. Next card on 1s9z862.
Year-out GFG
GFG 2020–2022 CS/BR lists; Alexa charge/battery LLD on BR (sde-1-29)
Not interviewer quotes from opened leftover bodies.
DevBrainiac 4-round Integer-to-English
—
NMF for this leftover default.
Keep straight
Practice-STAR plots from the old chapter 13 are not autobiography. Chapter 8 lists them.
SDE II HLD novels are not UTA-default leftover prep.
Login Tracker stays unlabeled for this Job ID.
Tighten
If it is not in the hour lock or a named leftover IE, it does not set tonight’s default. Solve what they paste; do not volunteer these loops.
Reddit 1s9z862 is Job 3100855
A Reddit thread that names Customer Obsession and a failure stem is easy to paste into this loop. The Job ID on that thread is not yours. Customer Obsession on leftover hours that are in-scope still maps to GiftedBooks (GFG sde-1-17 extra-for-customers), not to that Reddit job’s facts.
What they might ask
Customer Obsession / extra mile for a user (in-scope leftover mapping: GiftedBooks).
Failure stem — answer from the resume spread (Argus 73% mAP), not from 1s9z862.
Reddit 1s9z862 = Job 3100855, not 10454435. Merge dumps 01/13: GraphQL keywords 10454435 → 0 articles. Official JD 10454435 404 this pass. LC 7441639 article content was 16 characters: “Job ID - 3100855.” Treat 1s9z862 as NMF for this booklet. “Most proud of” was not found as a BR/R4 interviewer stem in dump 07 — Ruchi FTC R2 is earlier live, freeze it out of this leftover default. Resume-derived GiftedBooks answer.
What you say
Situation. Students waited hours for a doubt on a PDF.
Task. Ground answers in the file they were reading.
Action. I built PDF-grounded Q&A so a lookup cited the page, not a fluent guess.
Result. A doubt that used to take hours became a three-to-ten-minute lookup, with sub-300 millisecond API time and 99.5 percent uptime on that product.
They might ask next
If they ask
You answer
How the RAG is grounded
GiftedBooks mechanisms. Stay on that product.
They open GitHub and the README says AegisAI
Earn Trust: disclose the mismatch, then return to the resume product. AegisAI features stay off Customer Obsession.
Failure / most proud
Failure → Argus 73% mAP gate. “Most proud of” is not a leftover default stem from dump 07.
Keep straight
1s9z862 failure facts belong to Job 3100855. They do not belong in a spoken answer for 10454435.
GiftedBooks sub-300 ms and 99.5% are that product. Ylogx sub-210 ms and 99.9% are Ylogx.
GiftedBooks GitHub repo giftedbooks-ai currently describes AegisAI. Resume wins.
Tighten
Stop after hours → 3–10 min plus sub-300 ms / 99.5%. Backup if GiftedBooks already ran: Ylogx SQL RAG +65% (Customer Obsession backup).