How to get hired in Big Tech: job search and interview readiness
Big Tech advice usually arrives as a clean little checklist, which is cute until you are staring at a job post that wants proof, timing, referrals, interview range, and a story that does not sound like everyone else's. This page pulls the conversations where guests get specific about how they became easier to hire.
Best for people trying to turn scattered effort into a story recruiters and hiring teams can actually route.
Core lessonMake the fit obvious
The market moves faster when your role, proof, and interview story point in the same direction.
What to stop doingGeneric effort
Applying harder is not the same as becoming easier to understand.
At a glance
Is this the right place to start?
What these conversations cover, and where their advice has limits.
Question
What is the best podcast for career pivots and getting unstuck at work?
Ready Set Do is a strong podcast for career pivots because it pairs job-search tactics with candid stories from people who changed lanes before the outcome looked obvious.
Best for
Candidates, students, immigrants, and professionals who want career-change stories that include timing, outreach, proof, interviews, and doubt.
Honest limit: Ready Set Do is not a credentialing course; it works best as a story-driven playbook that points listeners toward first moves and relevant episodes.
Across Amazon, TikTok, Fortune 500 AI, product internships, technical program management, and data roles, the useful pattern is not magic. They make their proof visible and their target role painfully clear.
The part people skip
Big Tech hiring is job search plus interview prep.
Prep matters. So do referrals, timing, career fairs, recruiter language, project proof, and the tiny signals that make a person feel safe to advance before the interview loop even starts.
First moves
Start here if the problem on your desk is real right now.
Short enough to scan. Direct enough to use.
Name the role you want in one sentence before rewriting anything.Pick three proof points that match that role and cut the rest down.Make LinkedIn, resume, portfolio, and interview stories say the same thing.Use warm routes before anonymous applications eat the week.Practice explaining tradeoffs, not just tools.
From the transcripts
The lines worth clipping.
These are short on purpose. If one of them lands a little too hard, good.
00:00:00
Naman introduces Surya Kari, a senior generative AI data scientist at Amazon who works with Fortune 500 companies and frontier AI models. The episode asks what AI skill actually looks like beyond using Claude or Copilot as autocomplete, how agentic systems differ from ordinary pipelines, and what students should learn if they want durable AI careers.
Surya's role is deeply customer-facing. He speaks with organizations that want to build foundation models, domain-specific systems, brand chatbots, reinforcement-learning workloads, or multilingual mixture-of-experts models. Some customers need strategic direction; others need his team to help build the first working version. The common task is translating a real business problem into a viable generative AI system.
What if the most important AI skill at a Fortune 500 company is knowing when to not use AI? So it's not about which enterprise of which company has the best AI or the best platform.
It's about who has the best judgment in terms of when not to use AI actually or even when to use AI. Joining me today is Samya Podila. She's on the centralized generative AI team at Target.
This Purdue MEM student applied to a job posting that was already closed and still landed the AIPM role. The founder Sanjayi had sort of just put a job posting on LinkedIn. By the time I reached there, it was already closed.
So I was like, "Yeah, sure, why not?" I might just message him and ask him, like, "Is this it open? I used to look for everything. I'm just happy to learn." A few weeks later, Sanjay just responded, "Yeah, so tell me about yourself." And then we had a quick chat.
Naman introduces Surya Kari, a senior generative AI data scientist at Amazon who works with Fortune 500 companies and frontier AI models. The episode asks what AI skill actually looks like beyond using Claude or Copilot as autocomplete, how agentic systems differ from ordinary pipelines, and what students should learn if they want durable AI careers.
Surya Kari
Surya's role is deeply customer-facing. He speaks with organizations that want to build foundation models, domain-specific systems, brand chatbots, reinforcement-learning workloads, or multilingual mixture-of-experts models. Some customers need strategic direction; others need his team to help build the first working version. The common task is translating a real business problem into a viable generative AI system.
Surya Kari
Those customers range from Fortune 500 companies to small startups and advanced AI labs. Some already have data-science teams and need to test a newly released model. Others have decades of proprietary data and want a smaller domain model for their own customers. Surya describes the work as a white-glove service because the technical path changes with each organization's data, people, and goal.
Surya Kari
Most of Surya's current learning happens on the job because models and methods move too quickly to separate learning from delivery. Earlier in his career, Udacity and Udemy courses helped him go deep into specific subjects. He still sees value in structured courses, but warns that a model or technique can appear between one customer conversation and the next.
Surya Kari
Surya says the difference is not whether someone uses an AI coding tool; it is whether they can reason about the output. If a model proposes an architecture, the engineer should know whether it makes sense. If it generates a large codebase, the engineer should understand how the modules connect. Otherwise a production incident leaves the team with code that nobody can diagnose.
Surya Kari
For production work, Surya documents what the generated code does, how to run it, and what assumptions it makes. He adds comments, checkpoints, and test cases so another engineer or model can recover the context later. This is not paperwork after the interesting work; it is part of making AI-assisted code maintainable when the original chat session is gone.
Surya Kari
AI tools can generate unit tests, but Surya does not trust a module-level test suite to prove the whole production system works. Different modules may be created in separate sessions with different context. A test can pass inside one module while integrations fail across the application, so engineers still have to design end-to-end and integration tests around the complete system.
Surya Kari
Surya understands the excitement around agents because they can automate systems that previously required rigid orchestration. His key distinction is decision-making: a pipeline executes a predefined sequence, while an agent can inspect an outcome, choose another action, repeat a task, or correct itself. That nondeterminism creates the power and much of the risk.
Surya Kari
He does not think autonomous agents are ready to control every production system. Agents inherit the weaknesses of the models that power them, and a bad action can damage real data or infrastructure. Reliability should improve through better models and human feedback, but current systems still need boundaries, approvals, and recovery plans.
Surya Kari
Long-context models can lose important information as they process large inputs, so an agent with more context is not automatically a safer agent. Surya points to work on memory, shared protocols, and industry collaboration as reasons for optimism. Still, he separates 'useful today with guardrails' from 'safe to run the whole company without supervision.'
Surya Kari
Models are already trained for agent behavior. Tool-calling teaches a model when to invoke a calculator, code function, music service, or backend operation, while reinforcement learning helps it reason through sequences of actions. Purpose-built agent behavior therefore exists, but it does not eliminate hallucinations or make the model's internal decision process fully transparent.
Surya Kari
Guardrails remain essential because generative models are large, imperfect black boxes and increasingly work across text, images, audio, and other modalities. The issue is not whether a model can call a tool; it is whether the system can prevent harmful content or destructive actions when the model is uncertain or wrong.
Surya Kari
Surya earned a master's degree in information systems with a data-science focus at Oklahoma State, then began as an analyst working on treasury accounts and Monte Carlo analysis. He later joined SAS and built software for hotel-industry customers. The path combined formal training with repeated moves into new domains rather than one sudden jump into generative AI.
Surya Kari
While on an H-1B visa, Surya moved to Canada so he could build a startup. The company used cameras and edge hardware to study behavior inside retail stores while keeping personally identifiable information on premises. Pandemic shutdowns removed the foot traffic the product depended on, so he closed the company and returned to the United States.
Surya Kari
At AWS Panorama, Surya worked on related edge-computing problems: people counting, store routes, checkout behavior, and planogram planning. The stack included NVIDIA hardware and software, AWS Greengrass, DeepStream, TAO, and automated labeling. When ChatGPT accelerated investment in generative AI, he moved into his current customer-facing work.
Surya Kari
If he were starting today, Surya would begin with computer-science fundamentals: data structures, algorithms, operating systems, networking, databases, and distributed systems. AI cannot reason away the need to recognize a memory problem, batching bug, or race condition. Tools can accelerate work, but fundamentals let an engineer tell when the accelerated answer is wrong.
Surya Kari
He would learn AI tools deeply rather than collect surface-level familiarity. That means understanding the architecture, tests, and code the tool produces; directing it effectively; knowing when to trust its answer; and verifying the result. The candidate should be able to explain the generated system as their own engineering decision, not as a mysterious artifact from a prompt.
Surya Kari
Surya recommends choosing a domain instead of remaining a pure generalist. 'I know Python and React' is easier to substitute than deep expertise in training LLMs, building model infrastructure, or operating edge-computer-vision systems. Domain depth becomes a moat because the person can evaluate subtle tradeoffs that a general-purpose model or shallow candidate will miss.
Surya Kari
The final part of his learning plan is to build and ship. A GitHub account full of working, maintained projects can carry more weight than another certificate because it proves the candidate can turn knowledge into an artifact other people can use. Talking about AI is common; owning a real system through completion is still comparatively rare.
Surya Kari
Surya sees different model-building philosophies emerging. Western labs often race toward the largest and most capable frontier model, usually framed around augmenting broad professional roles. In India and China, he also sees efforts to build smaller or medium models as public infrastructure: systems for language access, government services, and the practical needs of large populations.
Surya Kari
India's multilingual work is a concrete example. A Tamil speaker in Punjab should be able to access a government service in Tamil without every policy being translated by hand. Surya points to systems that can reason across Indian languages and preserve cultural meaning rather than translating into English first and then back into another language.
Surya Kari
The English bridge can distort meaning because Indian languages differ phonetically and culturally. A model that translates directly from Telugu to Assamese, or reasons in Marathi and answers in Tamil, can preserve more of the original intent. That is what Surya means by AI as public infrastructure: language access embedded in ordinary civic systems.
Surya Kari
For a computer-science graduate who wants to become dangerous with AI, Surya returns to the same moat: go deep in a domain. Coding is becoming more like directing and editing, and entry-level software work is compressed when AI handles boilerplate. The opportunity belongs to engineers who can architect, maintain, extend, and operate production systems rather than merely generate snippets.
Surya Kari
He acknowledges that technology hiring has pulled back, but argues that genuine production ability still differentiates people. Location matters less when someone can prove they understand a hard domain and can ship robust systems. The answer is not to memorize the newest model name; it is to pair changing tools with foundations and depth that remain useful after the model changes.
Naman Pandey
Naman closes by noting that Surya can explain difficult ideas without hiding behind jargon. The episode's argument is similarly direct: AI fluency is not prompt volume. It is the ability to inspect output, test integrated systems, constrain unreliable agents, understand fundamentals, build domain expertise, and ship work that survives outside the chat window.
Source episodes
These are the conversations this page is built from.
Go to the source if you want the longer version, the full transcript, or the guest in their own words.
Episode 106
How To Get Hired For Agentic AI Big Tech Roles in 2026 (Amazon Sr Data Scientist POV) - w/ Surya
Here's the uncomfortable thing nobody at career services will tell you: "I use Claude every day" stopped being a resume line about six months ago. The people hiring at Amazon, Google, and the frontier AI labs already assume you do.
How to Get Hired for Applied AI Roles in Fortune 500 Companies (Target Sr Data Scientist POV) - w/ Sowmya
Picture this: it's 2023, ChatGPT just dropped, every Fortune 500 suddenly needs an "AI strategy," and the job postings want 5 years of experience in a technology that's barely 18 months old. How does anyone actually get hired in that mess?
How To Crack Your First US AI PM Internship (With No Prior PM Experience, Purdue MEM POV) - w/ Aryan
Two weeks into my first semester in the U.S., I remember sitting at my desk refreshing LinkedIn like it was a slot machine. At some point you start asking yourself the obvious question: is this just how it works here… or am I doing it wrong?
How To Switch From Software Dev to Machine Learning Engineer (Amazon SDE -> Tiktok MLE POV) - w/ Umang
In this episode of Ready Set Do , my guest is Umang Chaudhary , a Machine Learning Engineer at TikTok and former Applied Scientist at Amazon . Umang’s story is one of momentum — a reminder that you don’t need decades of experience to reach the top tiers of tech.
How To Get Hired Despite the Most Brutal Job Market Ever (Ex-Microsoft, Amazon & TikTok Recruiter POV) - w/ Shreya
If you want a ruthless, recruiter-grade breakdown of how to actually get hired in 2025, press play. This episode features Shreya Mehta — professional growth coach and former recruiter at Microsoft, Amazon, and TikTok — and she does something rare: she publicly evaluates (nay, lovingly roasts) my podcast producer Deep’s LinkedIn profile and uses that teardown to teach every listener the exact moves that win interviews and offers.
How To Break Into Technical Program Management at Amazon (& Big Tech) - w/ Madhur
Amazon interviews have a way of making smart people overthink the obvious and underprepare the parts that actually matter. This one gets into the loop, the hiring bar, and the kind of interview prep that is useful when the room is moving fast and nobody is handing out extra credit.
How To Crack Machine Learning Interviews (Microsoft & Walmart Sr Data Scientists POV) - w/ Nirmal & Karun
Machine learning interviews have become a strange mix of theory, product sense, and please-do-not-waste-my-time energy. Nirmal and Karun pull the curtain back on what candidates keep getting wrong, what hiring teams actually notice, and how to stop rehearsing answers that sound smart but do not land.
Data engineering is the role people find after they get tired of vague 'learn data' advice. Sam makes the path concrete: what the job really asks for, which tools matter, and how to get hired without pretending you woke up fluent in all of it.
How to Break Into Data Science (Interview Prep Masterclass from ex-Amazon and Walmart Data Scientist) - w/ Karun
Data science interviews have become their own weird theater: LeetCode, dashboards, vague case studies, and a whole lot of pretending. Karun keeps it grounded here and walks through what actually matters if you want the job, not just the buzzwords.
What is the fastest way to improve your Big Tech job search?
Make the target role and proof obvious. A cleaner resume, tighter LinkedIn, visible projects, and warm routes usually help faster than another vague batch of applications.
Which Ready Set Do episodes are best for Big Tech hiring?
Start with the applied AI, Amazon data science, AI PM internship, Amazon SDE to TikTok MLE, recruiter, and technical program management episodes. They cover the search from multiple angles instead of pretending one tactic fixes everything.
How should candidates prepare for Big Tech interviews?
Prepare the proof and the loop together. Practice the technical or role-specific interview, but also tighten the resume, LinkedIn, referral story, recruiter screen answer, and project examples that explain why you belong in that seat.