2. Intro and how to speak

Spoken notes for Adarsh Vishwakarma. Java Live Code. Job 10454435 still none. R1 and R2 done. Use this at the start of R3 or R4. Metrics only from the Aug 2026 resume. Labels: Resume-derived IE-asked.

Ninety-second intro Resume-derived

Memorize the shape, not the paragraph. Current role → recent role with numbers → breadth → why Amazon as your pattern. Lead with IQVIA, not “final-year student.”

Hi, I’m Adarsh Vishwakarma. I’m a final-year Computer Science student at CUSAT, and I’m a software developer intern at IQVIA in Kochi. There I architected two systems: a LangGraph multi-agent Deep Research platform that searches and ranks across more than two hundred websites, and a Hybrid RAG agent on Azure AI Search — hybrid and semantic retrieval plus a GraphDB — that works through two-hundred-page BRDs and PDFs, with LangSmith tracing, evals, and generated test cases.

Before that I interned about a year at Ylogx on a full-stack AI BI product: NestJS and Postgres, a LangChain SQL-RAG chatbot that let non-technical users query in English, which is the plus 65 percent data-analysis productivity number on my resume. I put three-tier row-level security in the database, Redis on the bot path for minus 35 percent database latency, and I deployed CloudFront, ECS, Docker, Route 53 and an ALB. We held 99.9 percent uptime and sub-210 millisecond dashboard responses.

Outside internships I build products — a computer-vision SaaS, a VR learning suite, and an industrial PPE system where I moved YOLOv9 from 73 to 89 percent mAP. I was core software on Team Horizon; we placed 17th globally at the European Rover Challenge among more than eighty teams. I’ve also picked up hackathon wins, including first at CodeRecet.

What draws me to Amazon is the same habit those projects share: own something end to end, ship it, and measure whether it actually helped — at a scale I cannot build alone.

That is about 75–90 seconds if you do not rush. Pause after your name. Do not list StratifyLabs, GiftedBooks, Argus, and eight hackathons in the intro; those wait for “tell me about a project.”

If they already heard Horizon in a previous round, drop the rover sentence and keep IQVIA + Ylogx + one product metric (Argus 73 to 89, or GiftedBooks hours to 3–10 minutes).

Forty-five-second intro Resume-derived

I’m Adarsh, final-year CS at CUSAT, software intern at IQVIA building multi-agent research and Hybrid RAG in LangGraph. Before that I spent a year at Ylogx on a full-stack BI platform with SQL RAG and AWS at sub-210 milliseconds and 99.9 percent uptime. I also ship my own AI products, and I was on our Mars rover team — 17th globally at the European Rover Challenge. I want to bring that build-and-ship habit to a team working at real scale.

When you know the answer

  1. Restate the problem in one sentence. Ask 3–6 clarifications out loud (null, duplicates, in-place, n size, return vs mutate).
  2. Name brute DS and why it is too slow, then optimal DS and why.
  3. Say time and space before you type the hot loop.
  4. Write Java in Live Code. Narrate. Dry-run one example. Name one edge unprompted.
  5. Stop. “Does that match what you wanted, or should I go deeper on X?”

Racing to code without an approach reads junior. Talking for three minutes after a complete answer dilutes it. For LP, Situation and Task are short; Action is three concrete steps; Result is a resume metric, then silence. They will ask why. Answer the why. Do not restart the STAR.

Use “I” for steps you took. Use “we” only for a team score you could not produce alone (Horizon 17th is a team result; Ylogx RLS in Postgres is an I-designed-and-shipped claim).

When you do not know

Amazon scores Earn Trust here. A wrong confident answer is worse than a bounded guess.

  1. Do not bluff Kafka, Qdrant, or a missed SLA.
  2. Say the adjacent thing you did ship. “I have not operated Kafka. I have run async jobs on Celery and Redis. I would expect the delivery-guarantee trade-off to look similar, and I would verify it before I shipped.”
  3. Reason out loud for CS trivia. “I am not sure of the exact bytecode behavior. My best guess is X. I would compile it before I treated it as fact.”
  4. If they open an IDE on Java locals: uninitialized local variables do not default; they fail to compile. Instance fields default to 0 / false / null. That is Standard CS, not a resume story.
  5. Do not let silence run past about ten seconds without “let me think through this for a second.”

If they ask Kafka because igreaper’s loop did: that loop was OA-as-R1 and not your intern stack. Do not fake partitioning. Pivot to Redis cache + RLS, then offer to reason about queues in general.

Q. What is your current role? IE-asked

GFG 2025 Bar Raiser asked current role. Prince-style loops want three one-liners then one STAR.

I am a software developer intern at IQVIA in Kochi, April 2026 to present. I work on two research systems. The first is a LangGraph Deep Research graph: Firecrawl, Bing, DuckDuckGo, and Playwright in parallel, then a ranker across more than two hundred websites. The second is Hybrid RAG: Azure AI Search hybrid and semantic retrieval, a GraphDB, LangGraph choosing retrieval, LangSmith for evals and generated test cases on more than two-hundred-page BRDs. Production is Python FastAPI. Live Code for this loop is Java.

Before IQVIA I interned at Ylogx, November 2024 to October 2025: FastAPI, NestJS, Postgres, SQL RAG, three-tier RLS, Redis, CloudFront and ECS, 99.9 percent uptime, 40 percent faster reports. Before that I was Horizon core software for ERC 2024, 17th of more than eighty. If they want three numbers: IQVIA 200+ sites; Ylogx 99.9 percent; Horizon 17th. Then I go deep on one.

Example of stopping. I do not list StratifyLabs, GiftedBooks, Argus, and eight hackathons in the current-role answer.

If they ask whether IQVIA is Java: I did not ship this as a JVM service. I would wrap the same graph of nodes in Java services if that were the shop.