Ivan's Cool Obsidian Vault

Oracle Prep

4 min read

  • based off reddit posts 1, 2, they tend to ask leetcode easy/mediums specifically in strings, stack, bst, and linked lists so that’s what i need to really prep for in the next week.(maybe converting decimal to binary)
  • brush up on OOP(inheritance, encapsulation, polymorphism, etc…) and what that looks like in java and python

Oracle AI / SWE Intern Interview – 1 Week Prep Plan

Assumption: Interview in ~7 days
Daily commitment: ~5 hours
Goal: Be strong in RAG fundamentals, applied system design, and articulation, not ML theory.

High-Level Strategy

You do not need to:

  • Master ML theory
  • Read all of Designing Data-Intensive Applications
  • Learn every RAG framework

You do need to:

  1. Explain RAG end-to-end clearly
  2. Reason about simple AI system design
  3. Tie everything back to your IBM experience
  4. Speak confidently about tradeoffs

Clarity > breadth > buzzwords

  • 2h – YouTube (watch + notes)
  • 1.5h – Reading (selective)
  • 1h – Talk out loud / write explanations
  • 0.5h – Review + tighten answers

⚠️ Talking out loud is critical.

Day 1–2: RAG Fundamentals (Non-Negotiable)

Goal

Be able to explain RAG clearly and calmly without buzzwords.

YouTube (Watch in Order)

  1. Harrison Chase – “RAG Explained”
    Search: Harrison Chase RAG explained

  2. AssemblyAI – “How ChatGPT Retrieval Works”

  3. Pinecone or Weaviate – “Vector Databases Explained”

Focus on flow, not tools.

Must-Be-Able-To Explain

A RAG system ingests documents, chunks them, embeds them into a vector database, retrieves relevant chunks at query time, and injects them into the prompt so the LLM answers grounded in source data.

Breakdown:

  • Ingestion
  • Chunking (size tradeoffs)
  • Embeddings
  • Vector search
  • Prompt construction
  • Generation
  • Evaluation

Reading (Skim Only)

O’Reilly – Data Engineering Book

  • Data ingestion
  • Batch vs streaming
  • Data quality

Skip heavy infra chapters.

Day 3: RAG Failure Modes & Evaluation (High Signal)

YouTube

  1. “RAG Failure Modes” (LangChain / LlamaIndex talks)
  2. “Evaluating RAG Systems” (Weaviate / Pinecone)

Topics to Internalize

  • Hallucinations
  • Irrelevant retrieval
  • Stale documents
  • Latency

Strong Talking Points

  • Better chunking
  • Metadata filtering
  • Hybrid search (keyword + vector)
  • Golden question evaluation
  • Confidence thresholds / fallback responses

Day 4: Light System Design (Beginner-Friendly)

Mindset

They are not asking:

“Design Facebook”

They are asking:

“How would you build an AI assistant for customers?”

YouTube

  1. Gaurav Sen – System Design for Beginners

    • Watch only first 2–3 videos

  2. “Design an AI Assistant” walkthrough (any modern video)

Ignore:

  • Sharding
  • CAP theorem
  • Scaling math

Focus on:

  • Components
  • Data flow
  • Tradeoffs

Practice Design (Out Loud)

Design an AI assistant for Oracle Cloud customers

Include:

  • Chat UI (frontend)
  • Backend API
  • RAG pipeline
  • Vector database
  • LLM
  • Logging & monitoring
  • Security (customer data isolation)

Day 5: Map Everything to IBM Experience (Huge Advantage)

Write This Down

  • What problem you solved
  • How data flowed
  • What went wrong
  • What you’d do differently today

Practice 2-Min Explanation

They will ask:

“Tell me about the AI assistant you worked on at IBM.”

You want:

  • Structured
  • Calm
  • No rambling

Day 6: Mock Interview + Gap Filling

Answer These Out Loud

  • What is RAG?
  • How would you reduce hallucinations?
  • How would you design an AI assistant for customers?
  • What tradeoffs did you face at IBM?

Record yourself if possible.

Review

  • Where you ramble
  • Where you’re vague
  • Tighten explanations

Day 7: Light Review + Confidence

  • Rewatch 1–2 key videos
  • Skim notes
  • No cramming
  • Sleep well

Confidence matters.

What NOT to Do

  • ❌ Don’t try to master DDIA
  • ❌ Don’t memorize frameworks
  • ❌ Don’t grind LeetCode
  • ❌ Don’t panic about system design

What a Strong Answer Sounds Like

“At a high level, here’s how I’d approach it… here’s the tradeoff… here’s what I’d improve…”

What a Weak Answer Sounds Like

“I’d probably use LangChain and embeddings and transformers…”

Final Reminder

This interview is about applied reasoning, not academic depth.

You are polishing — not catching up from zero.

Graph View