Am I ready to be an FDE?

What companies hiring forward deployed engineers ask for, from their public postings and interview guides. Tick what you can already do. Each requirement links to the lesson that teaches it and the practice that proves it.

0 of 35 must-haves

0 of 13 skills that set you apart

Engineering foundations

You can write, ship and debug real software in an environment you did not build: Python, APIs, SQL, Git, the terminal, containers and other people's code.

Strong Python for everyday field workYou write Python scripts and small services that read files and JSON, call APIs, handle dates and bad rows, log what they did and run unattended. You can explain every line in a code review.
Must haveFreshers tooAsked by Anthropic, OpenAI, Palantir, AI startups, Indian AI startups
Python for field work Messy data: CSV, JSON, Excel exports and dates Code Lab: Weekly sales by salesperson Code Lab: Clean Raftaar's consignment export
A second language, usually TypeScript or JavaYou can read and make a small change in a TypeScript or Java codebase, for example a field on an internal web page or a fix in a customer's Java service, without needing someone to walk you through the build.
Sets you apartAsked by Anthropic, OpenAI, Palantir
Reading an unfamiliar codebase fast Prototypes and demos that win trust
HTTP APIs from the client's sideYou call other people's APIs correctly: auth headers and tokens, status codes, paging until the last page, rate limits with backoff, timeouts, and safe retries. You can read an API document and write a working client in an afternoon.
Must haveFreshers tooAsked by Indian AI startups
HTTP and APIs from the client's side Glue code that doesn't break: retries, timeouts and idempotency Code Lab: Page through the CRM without losing anyone Field Lab: Ask for API access and sample data Code Lab: Charge orders through a flaky payments API
SQL to explore and check customer dataYou can join, group and filter tables to answer a question about a customer's data, find duplicates and missing rows, and check that a sync or report adds up before you show it to anyone.
Must haveFreshers tooAsked by OpenAI, Palantir
Your first query: SELECT Joining tables GROUP BY and HAVING SQL Lab: Bengaluru's top 5 SQL Lab: Users nobody has paid SQL Lab: Payment methods with over 2,000 deliveries
Git and code review in someone else's repoYou branch, commit in small steps, write a pull request a stranger can review, follow the customer's rules for review and release, and can undo a bad change calmly.
Must haveFreshers tooAsked by OpenAI
Git and code review in someone else's repo Field Lab: Pull request for a fix in Kaveri Bank's repo
The terminal, cloud basics and containersYou are at home on a Linux box over SSH: logs, processes, disk, environment variables, network checks. You can package a service in a container, run it on a cloud account and explain what ports, secrets and storage it needs.
Must haveFreshers tooAsked by Palantir, Indian AI startups
The terminal, Linux and cloud basics Containers and deploying a service Field Lab: A runbook for the customer's on-call team
Full-stack building: a small frontend on a backendYou can put a usable screen on top of your backend, such as a review queue, a search page or an admin tool, so that real users can try your work instead of reading a terminal.
Sets you apartAsked by OpenAI, Palantir
Prototypes and demos that win trust Building in stages as requirements change Code Lab: A credit ledger for Dukaan, built in stages
Debugging in production through logs and metricsGiven only logs and a few metrics, you can narrow a failure to one step, form a guess, test it without making things worse and write down what you found. You add the logging you wish had been there.
Must haveFreshers tooAsked by Anthropic, Indian AI startups
Logs, metrics and debugging in production Fixing code someone else wrote Code Lab: Find the slow endpoints in a log file Code Lab: Fix the appointment reminder script
Reading and fixing unfamiliar code fastYou can find your way round a few hundred or a few thousand lines you have never seen, reproduce a bug with a small test, fix it without breaking three other things, and explain the cause.
Must haveFreshers tooAsked by Palantir
Reading an unfamiliar codebase fast Fixing code someone else wrote The re-engineering and coding rounds Code Lab: Fix the sync that doubles and drops shipments Code Lab: Fix the appointment reminder script Code Lab: Fix Raftaar's daily delivery report Code Lab: Fix a headcount tally that double counts

Applied AI

You can turn a general AI model into a dependable system for one customer: pick the right use case, ground it in their data, give it tools, prove its quality with evals, keep it safe and roll it out in stages.

LLM applications: prompts, structured output, tool callingYou can write a clear system prompt with examples, get output in a fixed JSON shape and check it, and let a model call a tool you defined. You know why the same prompt can give different answers and how to make it steadier.
Must haveFreshers tooAsked by Anthropic, OpenAI, AI startups, Indian AI startups
Prompt engineering techniques Getting structured data back Tool use: letting Claude call your code Is this a good AI use case?Prompt Lab: Clear, structured and with examples Prompt Lab: Get clean data out Prompt Lab: Define a tool
Judging whether a task suits AI at allBefore building, you can say whether a request needs a model, a rule, a report or a process change, what a mistake would cost, and what a cheaper first step looks like. You can tell a customer kindly that their AI idea is weak.
Must haveFreshers tooAsked by OpenAI
Is this a good AI use case?Finding the real problem behind the request Customer Room: "Use AI to reduce delays"Customer Room: Two HR systems that disagree Prompt Lab: Decide what to hand to AI Field Lab: Assess Raftaar's AI request
Answers over private data, with permissionsYou can build search and question answering over a customer's documents: chunking, embeddings, hybrid search, citations, and filters so each employee only sees what they are allowed to see. You can explain why a wrong chunk came back.
Must haveFreshers tooAsked by Indian AI startups
Answers over a customer's private data Retrieval: answering from your documents Mixing keyword and meaning search Take-home: Product search that understands how people type Prompt Lab: Answer only from the documents Customer Room: A policy assistant the compliance team will accept Mock interview: Design a company knowledge assistant Code Lab: Permission-aware retrieval for Kaveri Bank
Agents and tools inside real workflowsYou can give a model a small set of well-described tools, decide which actions it may take alone and which it only drafts for a person to approve, keep an audit log, and stop it looping or acting on bad input.
Must haveAsked by Anthropic, Indian AI startups
Agents and tools in customer workflows Agents: tools, feedback and when to use them Many tools, many turns Take-home: A CRM agent with tools and an approval step Customer Room: An agent that updates the sales CRM Prompt Lab: Rules for an agent Mock interview: Design AI customer support Code Lab: A tool router for Raftaar's delay desk agent
MCP servers for a customer's toolsYou can expose a customer system, such as a CRM or a ticket desk, to a model through an MCP server with narrow tools, clear descriptions, the customer's own permissions and logging of every call.
Sets you apartAsked by Anthropic
Building MCP servers for a customer's tools Building an MCP server What MCP is Prompt Lab: A tool for your MCP server Prompt Lab: Tool, resource or prompt
Evals that decide what shipsYou build a golden set from real customer examples, grade answers with code, rubrics or a model grader you have checked, report one clear score per failure type, and rerun the suite before every change goes live.
Must haveFreshers tooAsked by Anthropic, OpenAI, AI startups, Indian AI startups
Evals: golden sets, graders and regression suites Testing prompts with evals Measuring a RAG system Take-home: Product search that understands how people type Prompt Lab: Design an eval Code Lab: Shadow-mode report for Suraksha's claims Code Lab: An eval harness for Sanjeevani's ticket triage Field Lab: A system prompt for Sanjeevani's ticket triage
Guardrails and human reviewYou decide where a person checks the AI's work, route low-confidence cases to a review queue, mask personal data before it reaches a model, and defend against prompt injection in documents and messages.
Must haveAsked by Indian AI startups
Guardrails and human review Defending against prompt injection Keeping personal data out of AI Take-home: Read hospital bills with confidence and review Code Lab: Mask patient details before the model Customer Room: A KYC review queue that someone wants to automate
Cost and latency budgetsYou can estimate what a feature will cost per month and how long each answer takes at the customer's real volume, and bring both down with caching, smaller models for easy cases, shorter prompts and batching.
Sets you apartAsked by Indian AI startups
Cost and latency budgets Why AI apps charge by the word Prompt caching Code Lab: Monthly cost estimator with a budget alert Prompt Lab: Lay out a prompt for caching
Staged rollout and monitoring after launchYou switch AI on in steps: shadow mode, a small pilot, then wider use, with a go or no-go number at each step. After launch you watch quality, cost and drift, and hand over a system the customer's team can run.
Must haveAsked by Anthropic, OpenAI
Shadow mode, pilots and go-live After launch: monitoring, drift and handover Drift and retraining loops Code Lab: Shadow-mode report for Suraksha's claims Field Lab: Rollout plan for Kaveri Bank's policy assistant Field Lab: Handover note for Sanjeevani's IT team

Enterprise systems and security

You can work inside a large company's systems and rules: map what exists, connect to it safely, move data without losing rows, pass identity and security reviews and handle personal data lawfully.

Mapping a customer's systems in week oneBy the end of your first week you can draw the systems that matter to your project, who owns each one, how data moves between them and where it is entered by hand, and you have asked for the access you need.
Must haveAsked by Palantir
Learning a customer's systems fast Working with legacy systems Field Lab: Week-one summary to your own team from Sanjeevani Customer Room: Predicting truck breakdowns from workshop logs
Integration patterns: APIs, webhooks, files and queuesYou can pick between calling an API, receiving a webhook, picking up a nightly file or reading from a queue, and build it so repeats are ignored, signatures are checked and nothing is silently dropped.
Must haveFreshers tooAsked by Palantir, Indian AI startups
Integration patterns: APIs, webhooks, files and queues Glue code that doesn't break: retries, timeouts and idempotency Queues and background work Code Lab: Verify Setu's webhooks and ignore repeats Field Lab: Options memo: how to get delivery events Code Lab: Reconcile orders with billing
Data pipelines that never lose or double a rowYou can move data between systems every night with incremental loads, keys for de-duplication, safe reruns and a reconciliation report that proves the counts match.
Must haveFreshers tooAsked by Palantir
Data pipelines in and out of customer systems Change data capture and the outbox Case: syncing two HR systems Code Lab: Nightly discharge sync from the hospital system Code Lab: Fix the sync that doubles and drops shipments Take-home: Sync two HR systems and reconcile them Customer Room: Two HR systems that disagree
Identity: SSO, SAML, OIDC and service accountsYou can explain how a customer's single sign-on works, wire your app to it with OIDC or SAML, map their groups to permissions, and use a service account with the least access for background jobs.
Sets you apartAsked by Anthropic
Identity: SSO, SAML, OIDC and service accounts "Sign in with Google": logged in without a password
Deploying inside the customer's networkYou know what changes when software runs in the customer's private cloud network, on their own servers or with no internet: images you can carry in, settings without a console, logs you can only see through their team.
Sets you apartAsked by Anthropic
Deploying inside the customer: VPC, on-premises and air-gapped Containers and deploying a service Field Lab: A runbook for the customer's on-call team
Passing a customer's security reviewYou can answer a security questionnaire honestly and quickly: where data is stored and for how long, who can reach it, how it is encrypted, what the AI provider sees, and what happens in an incident. You know what SOC 2 and ISO 27001 reports are for.
Sets you apart
Security reviews and compliance The ten big security risks in AI apps Field Lab: Answer five questions from Kaveri Bank's security review
Personal data, the DPDP Act and data residencyYou treat names, phone numbers, health and financial details with care: collect only what the purpose needs, mask before sending to a model, keep data where the customer requires, and explain the data flow to a privacy officer in plain words.
Sets you apartFreshers too
Personal data, the DPDP Act and data residency Keeping personal data out of AI Data residency and the DPDP Act Code Lab: Mask patient details before the model Field Lab: Explain the data flow to Sanjeevani's privacy officer Customer Room: Triage for a hospital help desk
Knowing one industry wellYou know the main systems, rules and daily work of one industry, such as banking, insurance or healthcare, well enough to ask sharp questions in the first meeting and spot what a compliance team will worry about.
Sets you apartAsked by Anthropic, Indian AI startups
Case: a bank's policy assistant Case: reading insurance claim documents Case: triaging a hospital's support tickets Customer Room: A policy assistant the compliance team will accept Customer Room: Reading claim documents, and the demand for 100 percent Customer Room: A KYC review queue that someone wants to automate

Discovery and scoping

You can find the real problem behind a request, break a vague goal into parts you can build, agree how success is measured and scope a first project small enough to succeed.

Running a discovery callYou prepare, ask open questions one at a time, ask to be shown the real work, collect numbers with their source, find who owns the data and who signs off, promise nothing you cannot yet know, and send a recap within a day.
Must haveFreshers tooAsked by Anthropic, OpenAI
Running a discovery call Field Lab: Question list for a first call at Kaveri Bank Customer Room: Triage for a hospital help desk Customer Room: "Use AI to reduce delays"
Finding the real problem behind the requestWhen a customer asks for a chatbot or "AI for delays", you can restate the request as a problem with a person, a number and a cost, and check it against what you saw on the floor.
Must haveFreshers tooAsked by OpenAI, Palantir
Finding the real problem behind the request Field Lab: Restate Dukaan's chatbot request as a problem Customer Room: A chatbot for nursing college admissions Customer Room: An AI demand forecast for kirana stores
Decomposing a vague goalGiven a goal like "reduce delays", you can split it into users, causes, data, decisions and pieces of work, size each piece and say which one to start with and why, mostly without writing code.
Must haveFreshers tooAsked by OpenAI, Palantir
Decomposing a vague problem The decomposition round Field Lab: Decompose Kaam's onboarding goal and pick a slice Customer Room: Predicting truck breakdowns from workshop logs
Scoping a first project and writing it downYou pick a first project that can show a measured result in four to eight weeks, and write a one-page scope with the problem, the users, what is in and out, the success measure, the data, the risks and who signs off.
Must haveAsked by OpenAI
Scoping the first project Writing a scoping document Field Lab: One-page scope for Raftaar's gate delays Customer Room: Reading claim documents, and the demand for 100 percent
Mapping users, buyers, IT, security and legalIn week one you know who uses the system, who pays, who owns the data, who must approve going live and who could quietly block it, and you have a plan to bring each of them along.
Must haveAsked by OpenAI, Palantir
Mapping the people: users, buyers, IT, security and legal Field Lab: Stakeholder map for Kaveri Bank's complaints project Customer Room: A KYC review queue that someone wants to automate
Caring about the people who use what you buildYou sit with the nurse, the dispatcher or the claims reviewer, learn their day, and design so that your system saves them time instead of adding a screen. You can say how a technical choice changes their work.
Must haveFreshers tooAsked by Palantir
Finding the real problem behind the request A week in the field Customer Room: A chatbot for nursing college admissions Customer Room: Predicting truck breakdowns from workshop logs
Agreeing a baseline and a success measureBefore you build, you record today's number (late deliveries, hours per claim, wrong answers) with its source, and agree with the customer which number will show the project worked.
Must haveFreshers tooAsked by OpenAI
How FDE work is judged Scoping the first project Running a discovery call Customer Room: An AI demand forecast for kirana stores Code Lab: Shadow-mode report for Suraksha's claims

Working with customers

You can explain technical work to anyone, write updates people read, demo well, push back without losing trust, stay calm when things break and build relationships that last beyond one project.

Explaining technical work to non-engineersYou can explain what you built, what it cannot do and what you need from them to a CEO, a nurse manager or a compliance head in two minutes, without jargon, with one number and one decision.
Must haveFreshers tooAsked by Anthropic, OpenAI, Palantir
Explaining technical work to executives What a forward deployed engineer does Field Lab: Explain the FDE role to a worried parent Field Lab: Two minutes with Kaveri Bank's CEO about a delay
Clear written updates and documentsYour weekly update fits on one screen: what changed, the numbers, risks with owners and dates, and what you need. You also write short design notes, runbooks and handover documents that others use.
Must haveFreshers tooAsked by OpenAI, Indian AI startups
Status updates and written communication A week in the field Field Lab: Week-one summary to your own team from Sanjeevani Field Lab: Handover note for Sanjeevani's IT team Field Lab: Raftaar weekly update from raw notes
Demos that win trustYou can run a fifteen-minute demo on the customer's own examples, show a failure case honestly, and end with a clear next step. Your demo works on a slow network and has a fallback.
Sets you apartFreshers tooAsked by Indian AI startups
Prototypes and demos that win trust Take-home: Product search that understands how people type
Saying no and managing scopeYou can refuse a request, such as 100 percent accuracy or a feature outside the scope, offer two options with a recommendation, and keep the project on its agreed size without souring the relationship.
Must haveAsked by OpenAI
Saying no and pushing back Managing scope and expectations Customer Room: Reading claim documents, and the demand for 100 percent Customer Room: An agent that updates the sales CRM Field Lab: Refuse Dukaan's auto-discount request with an alternative Field Lab: Change request for three new document types at Suraksha
Staying calm in incidents and with upset customersWhen your system fails at the customer's busiest hour, you stop the harm first, tell people what you know and when you will update them, fix it, and write a blameless postmortem with actions.
Must haveAsked by OpenAI
When things break on site: incidents and postmortems Handling a frustrated customer Field Lab: Reply to Sanjeevani's angry head of operations Field Lab: Blameless postmortem: duplicate order confirmations at Dukaan
Building long-term customer relationshipsCustomers ask for you by name because you kept every promise, told them bad news early and spotted their next good project before they did.
Sets you apartAsked by Anthropic, Palantir
Mapping the people: users, buyers, IT, security and legal Explaining technical work to executives Customer Room: A KYC review queue that someone wants to automate
Working on site and travellingYou are ready to spend real time at customer sites. Many 2025 and 2026 postings mentioned travel of about 25 to 50 percent; ask each employer what it means for their team.
Must haveFreshers tooAsked by Anthropic, OpenAI, Palantir
A week in the field Offers, travel and where the career leads

Ownership and getting hired

You can be trusted alone at a customer: you drive unclear work to a result, learn fast, turn field work into product, and you can show all of it to a hiring manager with evidence.

High agency in unclear situationsWhen nobody tells you what to do, you define a first step, get it agreed, deliver it and adjust. You can tell two true stories where you did this, with the result.
Must haveFreshers tooAsked by Anthropic, OpenAI, Palantir, Indian AI startups
How FDE work is judged Behavioural rounds and the project deep dive Field Lab: Map a job posting to your skills and a six-week plan
Owning a project from first conversation to productionYou have taken at least one project from talking to its users through building, measuring and running it, and you can walk someone through every decision and what you would change.
Must haveFreshers tooAsked by OpenAI, Palantir
What a forward deployed engineer does A portfolio that proves FDE skills Take-home: Product search that understands how people type Take-home: Clean messy exports into a weekly report
Turning field work into product and feedbackYou notice when the same custom work repeats at different customers, write it up as a reusable pattern or tool, and send the product and research teams specific, evidence-backed feedback.
Must haveAsked by Anthropic, OpenAI, Indian AI startups
Turning one-off field work into product Field feedback to product and research Field Lab: Propose turning a field script into a product feature
Learning new tools fast from their docsYou can pick up an unfamiliar library, API or customer system from its documentation in an hour, build something small with it, and ask good questions about what the docs leave out.
Must haveFreshers tooAsked by Anthropic, Palantir
The learning round Learning a customer's systems fast Code Lab: Page through the CRM without losing anyone
A portfolio with evidenceYou have one or two finished projects on realistic data, each with a README, before and after numbers, an eval or reconciliation, and a short recorded walkthrough. Your CV lines read as outcomes, not tool lists.
Must haveFreshers too
A portfolio that proves FDE skills An FDE CV and profile From fresher to FDE Take-home: Support ticket triage with a review queue Take-home: Clean messy exports into a weekly report Field Lab: Write the first screen of a portfolio README Field Lab: Rewrite five weak CV bullets for an FDE application
Ready for every round of the FDE interviewYou have practised a decomposition question out loud, a customer role-play, a re-engineering exercise, a take-home with a walkthrough, an LLM system design and your project deep dive.
Must haveFreshers tooAsked by OpenAI, Palantir
How the FDE interview works The take-home and its walkthrough The customer system design round The re-engineering and coding rounds Customer Room: Triage for a hospital help desk Mock interview: Design a company knowledge assistant Take-home: Product search that understands how people type
Years of customer-facing experience, for senior rolesSenior AI-lab postings in 2025 and 2026 asked for roughly 4 or more years, often with customer-facing work; Indian startup postings often asked for 1 to 4. Strong evidence can stand in for some years; treat these numbers as guides.
Sets you apartAsked by Anthropic, OpenAI, Indian AI startups
Switching into FDE from another role Finding FDE roles in India and abroad Field Lab: Ask an FDE you admire for a referral
Caring about safe, responsible AIYou can explain the risks of the systems you ship, own the work AI helped you make, and choose a safer design when a riskier one would be faster.
Sets you apartFreshers tooAsked by Anthropic
Guardrails and human review Diligence: owning what AI helps you make Prompt Lab: Own what AI helped you make

Your ticks are kept in this browser only. Postings change often: read the current ones for the roles you apply to.