FullStack
FullStack3 open roles

Junior Forward Deployed AI Engineer - AI/ML & Data Track

Remote (US)RemoteEntryFull-time

Posted yesterday

The role in short

Embed with enterprise clients to discover, architect, build, and operate production AI agents and retrieval systems.

About FullStack

FullStack is your AI-native engineering partner, built to turn AI capability into certainty. Most companies can demo AI; few can get it to production and prove what it produced. We close that gap with three capabilities, AI built into your products from the start, elite vetted talent who apply it, and transparent execution that shows value every step of the way. With now over 600 customers in North America, FullStack is one integrated partner for all your AI engineering needs. With FullStack, you move forward with confidence.

We’re Most Proud Of

  • Offering life-changing career opportunities to talented software professionals across the Americas.
  • Building highly-skilled software development teams for hundreds of the world’s greatest companies.
  • Having delivered hundreds of successful custom software solutions, which have positively impacted the lives and careers of millions of users.
  • Our 4.1-star rating on GlassDoor.
  • Our client Net Promoter Score of 68, twice the industry average.

The Position

Forward Deployed Engineers sit inside the client’s problem, not next to it. You embed with an enterprise team, help find the work that actually warrants AI, learn to architect the system, build it, and stay long enough to help make it survive contact with production.

This is a three-way role: consultant, operator, engineer.
You will grow into pressure-testing a business case with a COO or VP of Engineering while learning to diagnose why retrieval quality collapsed after a data refresh. The people who succeed here are fast learners who join discovery conversations on Monday and help ship what was scoped on Thursday.

On this track, your focus is data, retrieval, and model behavior: you help make agents trustworthy by getting the data, context, and evaluation right.

We work with regulated industries and Fortune 500 clients who have AI mandates, real constraints, and low tolerance for demos that don’t hold up.

How We Approach The Work

FDEs are trusted to help shape the solution, not only deliver it. Every engagement calls for sound judgment, which you will develop over time:

  • Reframes the request. Treats “we need an agent for X” as a hypothesis, not a spec.
  • Decomposes the problem. Outcome → process → decisions → data → systems → actors → constraints — before choosing technology.
  • Helps choose the intervention. Remove, simplify, traditional software, automation, LLM, RAG, agent, or multi-agent — and learns to defend why. Starts to recognize when the answer is a SQL query, not a vector database.
  • Contributes to architecture decisions. Learns to document trade-offs (ADRs), set success metrics, and raise concerns about scope that won’t deliver value.

What You'll Do

Discover.
Support workshops and process discovery with business, data, and engineering stakeholders. Help assess data readiness honestly. Learn to separate genuine AI problems from workflow problems wearing an AI costume.

Structure the process and the spec.
For any business process being automated, help map the current state, decision points, data inputs, and exceptions, then turn it into specifications that both people and AI systems can execute against reliably.

Architect & design.
Contribute to designing parts of agentic systems with emphasis on context and retrieval architecture — data sourcing, chunking, embeddings, vector and hybrid search, with exposure to reranking and permissions-aware retrieval — plus orchestration, MCP/tool integration, guardrails, and human-in-the-loop checkpoints.

Build.
Help ship working systems — agents, retrieval services, evaluation tooling, and the data flows that feed them. Prototypes that prove value in weeks, not slideware.

Engineer the delivery system.
Contribute to CI/CD for what you build, extended to AI: prompt and context versioning, learning to add eval suites that run in the pipeline.

Operate (Day 2).
Support the team after launch — eval harnesses, groundedness and retrieval metrics, observability and tracing, cost and latency management, drift, and supporting the handoff that lets the client’s team run it without you.

Consult.
Help prepare and present decisions to CTOs, VPs of Engineering, CDOs, and business leadership. Learn to quantify impact in their terms. Observe pre-sales scoping and proposal work when the deal calls for it.

What We're Looking For

Engineering foundation

  • About 1 year of hands-on experience across ML, data, and software engineering (flexible; internships, apprenticeships, and strong project work count)
  • Python-first, clean-code mindset (typed, tested, deployed projects — not notebooks only); basic familiarity with TypeScript / Node, .NET / C#, or Java is a plus
  • College graduate (bachelor’s degree) in Computer Science, Engineering, Data Science, or a related field
  • Relevant certifications in AI/ML, data, or cloud (e.g., Azure AI Engineer, AWS Machine Learning, Google Professional ML Engineer, Databricks, or equivalent)

Applied / operational AI

  • Has built LLM or agent projects, ideally exposed to production environments (internships, open-source, or capstone work count)
  • Working knowledge of retrieval and model behavior: embeddings, vector search, structured output, and prompting vs. retrieval trade-offs; eager to learn reranking, hybrid search, and model selection
  • Hands-on with agentic tooling: Claude Code or similar AI coding tools, agent frameworks, MCP, and tool/function calling; curiosity about multi-agent orchestration
  • Context and spec engineering: learns to design what the model sees — retrieval, context construction, specifications, process definitions, and memory — for reliable, context-efficient output
  • Exposure to Day 2 topics: evaluation design (golden sets, LLM-as-judge, groundedness), observability, cost and token economics, failure-mode analysis
  • Data and integration literacy: basic familiarity with ETL/ELT patterns, warehouses/lakehouses (Snowflake, Databricks, BigQuery), and integration options (APIs, event streams, iPaaS, MCP) — enough to understand them and contribute under guidance. Does not need to be an ETL/ELT specialist.

Consulting & business

  • Shows curiosity and early signs of the discovery behaviors above: business and personal drivers, constraints, stakeholder alignment, strategic questioning
  • Interest in growing into client conversations (workshops, proposal walkthroughs, tough questions) as you gain experience
  • Learning to translate technical decisions into business outcomes — accuracy, cycle time, cost per transaction, risk reduction, ROI
  • Strong written and verbal communication in English; comfortable learning in front of senior stakeholders

Strong Plus

  • Additional certifications beyond the required ones (e.g., Databricks, Snowflake, or cloud data/AI specialty certifications)
  • Public portfolio of work: GitHub projects, open-source contributions, or hackathon and capstone projects involving LLMs or agents
  • Academic or project experience in classical ML (forecasting, classification, anomaly detection) alongside LLM work
  • Internship or project experience with stakeholder-facing work, such as presenting demos or documenting solutions
  • Comfort with ambiguity, fast learning, and switching between projects; prior internship or agency exposure is a bonus

Nice To Have
We’re Seeking Engineers Who Are AI-enthusiastic In Their Workflow. Professional Experience Or Personal Project Work With The Following Is a Significant Plus

  • Cursor, Windsurf, or Google Antigravity.
  • Claude Code, Gemini CLI, Codex, or Copilot CLI.

Work Authorization Requirements

  • Applicants must be legally authorized to work in the United States. FullStack Labs is unable to sponsor or take over sponsorship of an employment visa for this position.
  • FullStack Labs is an equal opportunity employer. We do not discriminate on the basis of citizenship status, immigration status, or national origin, or any other protected characteristic. FullStack Labs participates in E-Verify.

What We Offer

  • Competitive Salary.
  • Paid Time Off (vacation, sick leave, parental leave, holidays).
  • 100% remote work.
  • The ability to work with leading startups and Fortune 500 companies.
  • Health, dental, and vision insurance.
  • 401(k) w/ 4% match.
  • Ample opportunity for career advancement.
  • Continuing education opportunities.

FullStack is proud to be an equal opportunity workplace. We are committed to equal employment opportunity regardless of race, color, ancestry, religion, sex, national origin, sexual orientation, age, citizenship, marital status, disability, gender identity, or Veteran status.

Prep for this interview

5 questions likely to come up at FullStack, each matched to a line in this posting.

  1. 1Coding
    They wrote: “Python-first, clean-code mindset (typed, tested, deployed projects — not notebooks only); basic familiarity with TypeScript / Node, .NET / C#, or Java is a plus”

    Here's a CSV export from a customer. Write a script that cleans it and loads it into our API.

    How to answer

    Why they ask: Very close to real FDE work. They watch how you handle bad rows and failures.

    1. Read with a proper CSV parser; check the header first
    2. Validate each row and collect errors, not crash on the first one
    3. Send in batches with retries and a clear report at the end
    4. Write one or two quick tests for the parsing

    Tip: Talk through edge cases out loud: encodings, missing fields, duplicates, rate limits.

  2. 2Discovery and scoping
    They wrote: “Support workshops and process discovery with business, data, and engineering stakeholders. Help assess data readiness honestly. Learn to separate genuine AI problems from workflow problems wearing an AI costume.”

    A customer says “we want AI in our claims process.” What do you ask in the first meeting?

    How to answer

    Why they ask: Most FDE work starts this vague. They want to hear you find the real problem before you pick a tool.

    1. Map the current process: who touches a claim, in what order, with which systems
    2. Find the slowest or most error-prone step and ask what it costs them
    3. Agree on one measurable goal (for example time per claim, or error rate)
    4. Ask what data you can access, and who signs off on security

    Tip: Ask about the work, not the technology. Who does it today, how long it takes, what a mistake costs, and how they'd know it got better.

  3. 3LLMs, RAG and agents
    They wrote: “Contribute to CI/CD for what you build, extended to AI: prompt and context versioning, learning to add eval suites that run in the pipeline.”

    How do you know an LLM feature is good enough to ship?

    How to answer

    Why they ask: Evals are what separate a demo from production. Many roles list them by name.

    1. Write 50–200 real examples with expected outcomes, with customer experts
    2. Agree on a pass bar and the errors that are never acceptable
    3. Score automatically where possible; review a sample by hand
    4. Re-run on every prompt or model change, and watch live traffic after launch

    Tip: Describe a test set built with the customer, a pass bar agreed up front, and checks that run on every change.

  4. 4Cloud, deployment and security
    They wrote: “Relevant certifications in AI/ML, data, or cloud (e.g., Azure AI Engineer, AWS Machine Learning, Google Professional ML Engineer, Databricks, or equivalent)”

    How would you debug a deployment that works in staging but fails at the customer's site?

    How to answer

    Why they ask: Customer environments differ in ways you can't see. This tests method under pressure.

    1. Compare versions, config and environment variables
    2. Check network rules, proxies and certificates (common in enterprise)
    3. Look at resource limits and permissions
    4. Add the missing check to your install steps afterwards

    Tip: List the usual suspects and how you'd check each without full access.

  5. 5Data and SQL
    They wrote: “Helps choose the intervention. Remove, simplify, traditional software, automation, LLM, RAG, agent, or multi-agent — and learns to defend why. Starts to recognize when the answer is a SQL query, not a vector database.”

    Write a query that finds each customer's first and most recent order.

    How to answer

    Why they ask: A quick check that you're comfortable with grouping, joins and window functions.

    1. Clarify the tables and what counts as an order (cancelled ones?)
    2. Use MIN/MAX with GROUP BY, or ROW_NUMBER() if you need the whole row
    3. Handle ties on the same timestamp, and customers with no orders
    4. Explain how you'd test it on a few known customers

    Tip: Say what you'd check in the data before writing the query: duplicates, time zones, nulls.

Not the one? Get the next one by email

New roles land here every day. Pick what you want to hear about.

FDE Info

Want the job, not just the posting? Learn how FDEs work.

FDE Info is our free weekly newsletter for forward deployed engineers: tools, field notes from real deployments, and how FDE teams hire and pay.

Free. One email a week. Unsubscribe in one click.

More at FullStack

Similar roles at other companies

FullStack

Remote

Apply