Forward-Deployed AI Engineer (Contract)
Posted 11 days ago
The role in short
Embed with financial services clients to design, build, and deploy AI agents, MCP servers, and workflow automations within their existing technology stack.
Vaughn Harvey & Co is a boutique AI advisory and delivery firm. We act as fractional Chief AI Officer and transformation partner to financial and professional services firms in New York, bringing AI agents into the daily work of their teams.
The role
You'll be the engineer in a small pod that goes into one business function at a time, redesigns a workflow, builds it, trains the team and hands it over. You build what the business can't build itself: connectors, MCP servers, agents and automations with human checkpoints, inside the client's own stack and development process, so it keeps running after you leave.
What you'll do
- Build and integrate MCP servers, the client's own and third-party (Microsoft 365 and Dynamics, Google Workspace, Salesforce, Snowflake, Zendesk), with the right permissions.
- Build a function's first AI workers: triage, drafting and routing agents, with a person in the loop wherever the rules require it.
- Write evals and monitoring so agents stay within bounds and failures surface.
- Handle sensitive and regulated data correctly from the first commit.
- Document what you build so the client's team can run it without you.
What you bring
- 4+ years of software engineering, with at least a year building on LLM platforms; hands-on with Claude skills, MCP servers and tool use.
- Strong Python or TypeScript; comfortable with OAuth, REST APIs and JSON schemas.
- A security-first habit with personal and regulated data. Financial services, healthcare or insurance experience is a plus.
- A product mindset and clear writing: an agent's instructions are the product.
- Comfort working inside a client's stack, with business owners in the room.
Nice to have: Azure OpenAI and Dynamics/Dataverse, Snowflake Cortex, Glean or Salesforce Agentforce; published skills, plugins or MCP servers.
How we hire
A 30-minute conversation, a walkthrough of something you've built, a short paid practical, then a conversation with the client.
To apply
Send your resume and a link to something you've built (GitHub, an MCP server, a skill), with your hourly rate and earliest start.
Vaughn Harvey & Co welcomes applicants of every background.
Pay: $85.00 - $140.00 per hour
Expected hours: 40.0 per week
People with a criminal record are encouraged to apply
Application Question(s):
- What hourly rate are you looking for, and would you work 1099 or through your own company? When could you start?
- Please share a link to something you've built on an LLM platform (GitHub, an MCP server, a Claude skill or plugin) and one line on what it does.
Ability to Commute:
- New York, NY 10017 (Preferred)
Work Location: Hybrid remote in New York, NY 10017
Prep for this interview
4 questions likely to come up at Vaughn Harvey & Co, each matched to a line in this posting.
- 1Integrations and APIs
They wrote: “Strong Python or TypeScript; comfortable with OAuth, REST APIs and JSON schemas.”
How would you set up single sign-on for a large customer?
How to answer
Why they ask: SSO (SAML or OAuth/OIDC) is a common first task in enterprise deals and a common source of delays.
- Confirm their identity provider (Okta, Entra ID, Google) and protocol
- Exchange metadata and test in a staging tenant first
- Map user groups to roles in your product
- Plan for user removal: what happens when someone leaves the company
Tip: Mention the people involved, and the protocol second. The customer's identity team usually controls the timeline.
- 2LLMs, RAG and agents
They wrote: “Write evals and monitoring so agents stay within bounds and failures surface.”
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.
- Write 50–200 real examples with expected outcomes, with customer experts
- Agree on a pass bar and the errors that are never acceptable
- Score automatically where possible; review a sample by hand
- 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.
- 3Cloud, deployment and security
They wrote: “A security-first habit with personal and regulated data. Financial services, healthcare or insurance experience is a plus.”
The customer's security team sends a 200-question review. How do you handle it?
How to answer
Why they ask: Security reviews often block deals. FDEs help get through them.
- Start from your company's standard answers (SOC 2 report, past reviews)
- Answer clearly; don't oversell
- Bring in your security team for anything you're unsure of
- Keep the new answers for the next customer
Tip: Reuse past answers, flag honest gaps, and offer a call for the hard questions.
- 4Coding
They wrote: “4+ years of software engineering, with at least a year building on LLM platforms; hands-on with Claude skills, MCP servers and tool use.”
Build a small in-memory cache. Then add expiry. Then a size limit.
How to answer
Why they ask: Staged exercises show how your code holds up when requirements change mid-interview, like at a customer.
- Start with get and set on a dictionary, plus tests
- Store an expiry time with each value and check it on read
- Track use order and evict the least recently used entry when full
- Say what you'd change for many threads or many machines
Tip: Keep the first version simple and well tested so the next stage is a small change.
Not the one? Get the next one by email
New roles land here every day. Pick what you want to hear about.
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.