# Forward Deployed Engineer — Take-Home Exercise

## Context

Docktape runs a cloud cost optimization service. Prospective customers apply through a web form on our website to learn more. Right now, every submission lands in a shared inbox with zero context — someone on the sales side has to manually look the company up, and manually check whether there's any reason *not* to engage, before deciding whether it's worth a call.

The current form collects:

- Name
- Email
- Company name
- Website

You're welcome to extend these fields if it helps (e.g. job title, company size band) — your call, and you should justify it.

## How we expect you to work

We expect you to lean heavily on AI coding assistants — Claude Code, Cursor, Copilot, whatever you prefer — to actually build this. That's the point of the exercise. We're not grading whether you typed every line by hand; we're grading whether you can direct AI tooling to ship a real, working, sensibly-scoped prototype quickly. Tell us in your README which tools you used and how.

There's no required infrastructure or cloud provider. A local script, a notebook, a small app — anything that's real, runnable, and demoable is fine. Don't spend time on deployment unless it genuinely helps you demo the thing.

## What the prototype must do

1. **Accept a lead submission.** Build a tiny form, or just trigger the pipeline with a sample payload — either is fine. What matters is a real, working pipeline you can demo, not a mockup.

2. **Research the company.** Use whatever public data sources you think are reasonable (the submitted website, public search, an enrichment API, an MCP server you wire up yourself, etc. — your choice, justify it), and use an LLM to turn that research into a short, useful summary.

3. **Run a quick compliance/legal screening agent.** Before a lead gets marked sales-ready, check it against the "do-not-engage" list below. This should be a real check an agent performs (semantic/fuzzy matching, reasoning about partial matches), not a hardcoded `if` against exact strings. Flag any match or plausible near-match, with reasoning.

   **Sample do-not-engage list** (fictional, for this exercise):
   - Known competitors: *CloudTrim Inc*, *SpendWise Cloud*, *RightSize Cloud Co*
   - Any company headquartered in a country on a standard sanctions/export-control list (e.g. OFAC's SDN-related country list) — you can reason about this generally, no need for a real-time sanctions database

4. **Produce a fit score.** We care most about company size and any signal you can reasonably infer about how much the company likely spends on cloud infrastructure. How you define and calculate "fit" is up to you — be ready to explain your reasoning.

5. **Write the result to a tracker.** Append/update a row in an Excel (or Google Sheets) tracker with the company info, fit score, compliance flag, and a short summary — this should be a real spreadsheet a sales rep could open and skim down, not a database table only you can read.

6. **Notify someone.** Send the same summary as a Slack message or email to a (fake/test) sales rep.

## What to submit

- A link to your code (GitHub repo or zip)
- A short README covering: the AI coding tools you used and how, your data-source choice and why, how you defined "fit," how your compliance check handles fuzzy/partial matches, and any other assumptions
- Proof it runs: the resulting spreadsheet row(s), plus a screenshot or recording of the Slack/email output, for at least one lead — ideally one that's a clean "good fit," one borderline case, and one that should get flagged by the compliance check
- *(Optional)* a short Loom/video walkthrough, if you'd rather talk us through it than write it up

## Time & logistics

This should take a few focused hours of work, not days. You'll have a couple of days (through the weekend, if that's easier) to fit it around your schedule — there's no expectation you use the whole window.

## What we're not looking for

A polished UI, real legal expertise, a perfect sanctions database, or a "correct" fit formula. We're interested in how you reason through a real, slightly-underspecified business problem, how you use AI tooling to move fast, and whether you ship something that actually works end to end.