Job Description
UPWORK JOB POSTING
Fractional AI & HubSpot Automation Developer
Project | Part-Time | Remote
HubSpot API
Make.com / n8n
Claude / OpenAI API
Python or JavaScript
Who We Are
We are a B2B SaaS company. We run HubSpot Marketing Hub Professional and Sales Hub Professional as our CRM and revenue platform.
What We Are Building
Project 1 — Form Submission Classification Agent (Start Here)
This is the first project and the one we want to move on immediately. We have a serious data quality problem caused by non-sales submissions polluting our HubSpot CRM — job applicants, existing customers seeking support, and internal staff using our sales forms incorrectly. This corrupts our MQL counts, and distorts our reporting.
What we need built: a webhook-triggered agent that intercepts every HubSpot form submission, sends the data to the Claude API for classification, and routes it based on the result — genuine sales leads pass through to CRM as normal, everything else gets redirected or quarantined before it touches our data.
Deliverables: working Make.com or n8n workflow, Claude API prompt with structured JSON output, HubSpot workflow integration, and a simple dashboard or log showing classification decisions for audit purposes. We want to review every classification for the first two weeks before switching to fully automated mode.
The Broader Roadmap — What Comes After
We are sharing our full roadmap to the right candidate Strong performance on Project 1 leads directly into these:
Skills Required — All Four Are Mandatory
We are not looking for a generalist. We need someone with demonstrated hands-on experience in all four of these areas:
HubSpot API
You have built real integrations using the HubSpot API — not just used HubSpot as a user. You understand webhooks, contact properties, workflows, and the data model.
Make.com or n8n
You build complex multi-step automations, not beginner workflows. You know how to handle errors, retries, and conditional logic.
LLM / AI APIs
You have built production systems using Claude, OpenAI, or similar APIs. You know how to write reliable structured prompts, handle JSON outputs, manage token costs, and build error handling.
Python or JavaScript
Some of what we are building requires custom code. Zapier-only experience is not sufficient.
What a Strong Candidate Looks Like
They have built a HubSpot integration that involved external API calls — not just internal HubSpot automations
Has used Claude or OpenAI API to build something that produced structured JSON output used downstream in a workflow
Can explain the difference between a prompt that reliably produces structured output and one that does not
Asks clarifying questions before estimating — does not say yes to everything immediately
Communicates clearly in English and can translate business requirements into technical solutions independently
Is comfortable working async and providing brief weekly progress updates without being micromanaged
Engagement Structure
Type
Project-based to start, with strong potential for ongoing retainer
Hours
Flexible — scoped per project, we are not prescribing hours upfront
Location
Open to all regions globally — English communication required
Duration
First project is defined below. If the working relationship is strong, we expect this to become a long-term engagement across a multi-project roadmap
Communication
Async-first with availability for a weekly or bi-weekly sync call
First Project
Form Submission Classification Agent (full brief provided to shortlisted candidates)
Rate
Please state your hourly
How to Apply — Read This Carefully
We will not respond to generic applications. To be considered, your proposal must answer the following three questions specifically:
Question 1: Describe a HubSpot integration you built that involved an external API or webhook. What was the trigger, what did it do, and what was the business outcome?
Question 2: Describe a system you built using Claude, OpenAI, or another LLM API where the AI returned structured data (e.g. JSON) that was used downstream in an automation. How did you handle cases where the AI returned malformed or unexpected output?
Question 3: Our first project is a form submission classifier. Every time a HubSpot form is submitted, we want an AI agent to read the submission and decide if it is a genuine sales lead, a support request, a job applicant, or internal misuse — then route it accordingly. In 3-5 sentences, describe how you would architect this. What tools would you use and why?
Applications that do not answer all three questions will not be reviewed. We are looking for a long-term partner, not the fastest or cheapest proposal.
Paid Test Project
Before committing to Project 1, we will ask shortlisted candidates to complete a small paid test. This will be approximately 3-5 hours of work compensated at your stated rate. The test will involve building a Make.com or n8n workflow that receives a simulated HubSpot form submission via webhook, sends it to the Claude API with a classification prompt, and returns a structured JSON result. We will provide sample form data and evaluate both the technical implementation and the quality of the classification prompt you write.
This step is non-negotiable. It protects both sides — you get paid for your time, and we get a real signal of how you work before committing to a larger project.
Please Do Not Apply If...
Your HubSpot experience is limited to using it as a CRM user — we need API-level work
Your only automation experience is with Zapier basic workflows
You have not actually built with an LLM API before
You are not available for an ongoing retainer arrangement
You cannot communicate clearly in written English
Apply tot his job
Apply To this Job